Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
58862699 | 2 | * linux/fs/jbd2/transaction.c |
470decc6 DK |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 | |
5 | * | |
6 | * Copyright 1998 Red Hat corp --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Generic filesystem transaction handling code; part of the ext2fs | |
13 | * journaling system. | |
14 | * | |
15 | * This file manages transactions (compound commits managed by the | |
16 | * journaling code) and handles (individual atomic operations by the | |
17 | * filesystem). | |
18 | */ | |
19 | ||
20 | #include <linux/time.h> | |
21 | #include <linux/fs.h> | |
f7f4bccb | 22 | #include <linux/jbd2.h> |
470decc6 DK |
23 | #include <linux/errno.h> |
24 | #include <linux/slab.h> | |
25 | #include <linux/timer.h> | |
470decc6 DK |
26 | #include <linux/mm.h> |
27 | #include <linux/highmem.h> | |
e07f7183 | 28 | #include <linux/hrtimer.h> |
47def826 | 29 | #include <linux/backing-dev.h> |
44705754 | 30 | #include <linux/bug.h> |
47def826 | 31 | #include <linux/module.h> |
470decc6 | 32 | |
343d9c28 TT |
33 | #include <trace/events/jbd2.h> |
34 | ||
7ddae860 | 35 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh); |
de1b7941 | 36 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh); |
7ddae860 | 37 | |
0c2022ec YY |
38 | static struct kmem_cache *transaction_cache; |
39 | int __init jbd2_journal_init_transaction_cache(void) | |
40 | { | |
41 | J_ASSERT(!transaction_cache); | |
42 | transaction_cache = kmem_cache_create("jbd2_transaction_s", | |
43 | sizeof(transaction_t), | |
44 | 0, | |
45 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY, | |
46 | NULL); | |
47 | if (transaction_cache) | |
48 | return 0; | |
49 | return -ENOMEM; | |
50 | } | |
51 | ||
52 | void jbd2_journal_destroy_transaction_cache(void) | |
53 | { | |
54 | if (transaction_cache) { | |
55 | kmem_cache_destroy(transaction_cache); | |
56 | transaction_cache = NULL; | |
57 | } | |
58 | } | |
59 | ||
60 | void jbd2_journal_free_transaction(transaction_t *transaction) | |
61 | { | |
62 | if (unlikely(ZERO_OR_NULL_PTR(transaction))) | |
63 | return; | |
64 | kmem_cache_free(transaction_cache, transaction); | |
65 | } | |
66 | ||
470decc6 | 67 | /* |
f7f4bccb | 68 | * jbd2_get_transaction: obtain a new transaction_t object. |
470decc6 DK |
69 | * |
70 | * Simply allocate and initialise a new transaction. Create it in | |
71 | * RUNNING state and add it to the current journal (which should not | |
72 | * have an existing running transaction: we only make a new transaction | |
73 | * once we have started to commit the old one). | |
74 | * | |
75 | * Preconditions: | |
76 | * The journal MUST be locked. We don't perform atomic mallocs on the | |
77 | * new transaction and we can't block without protecting against other | |
78 | * processes trying to touch the journal while it is in transition. | |
79 | * | |
470decc6 DK |
80 | */ |
81 | ||
82 | static transaction_t * | |
f7f4bccb | 83 | jbd2_get_transaction(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
84 | { |
85 | transaction->t_journal = journal; | |
86 | transaction->t_state = T_RUNNING; | |
e07f7183 | 87 | transaction->t_start_time = ktime_get(); |
470decc6 DK |
88 | transaction->t_tid = journal->j_transaction_sequence++; |
89 | transaction->t_expires = jiffies + journal->j_commit_interval; | |
90 | spin_lock_init(&transaction->t_handle_lock); | |
a51dca9c | 91 | atomic_set(&transaction->t_updates, 0); |
8f7d89f3 JK |
92 | atomic_set(&transaction->t_outstanding_credits, |
93 | atomic_read(&journal->j_reserved_credits)); | |
8dd42046 | 94 | atomic_set(&transaction->t_handle_count, 0); |
c851ed54 | 95 | INIT_LIST_HEAD(&transaction->t_inode_list); |
3e624fc7 | 96 | INIT_LIST_HEAD(&transaction->t_private_list); |
470decc6 DK |
97 | |
98 | /* Set up the commit timer for the new transaction. */ | |
b1f485f2 | 99 | journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires); |
470decc6 DK |
100 | add_timer(&journal->j_commit_timer); |
101 | ||
102 | J_ASSERT(journal->j_running_transaction == NULL); | |
103 | journal->j_running_transaction = transaction; | |
8e85fb3f JL |
104 | transaction->t_max_wait = 0; |
105 | transaction->t_start = jiffies; | |
9fff24aa | 106 | transaction->t_requested = 0; |
470decc6 DK |
107 | |
108 | return transaction; | |
109 | } | |
110 | ||
111 | /* | |
112 | * Handle management. | |
113 | * | |
114 | * A handle_t is an object which represents a single atomic update to a | |
115 | * filesystem, and which tracks all of the modifications which form part | |
116 | * of that one update. | |
117 | */ | |
118 | ||
6d0bf005 | 119 | /* |
28e35e42 | 120 | * Update transaction's maximum wait time, if debugging is enabled. |
6d0bf005 TT |
121 | * |
122 | * In order for t_max_wait to be reliable, it must be protected by a | |
123 | * lock. But doing so will mean that start_this_handle() can not be | |
124 | * run in parallel on SMP systems, which limits our scalability. So | |
125 | * unless debugging is enabled, we no longer update t_max_wait, which | |
126 | * means that maximum wait time reported by the jbd2_run_stats | |
127 | * tracepoint will always be zero. | |
128 | */ | |
28e35e42 TM |
129 | static inline void update_t_max_wait(transaction_t *transaction, |
130 | unsigned long ts) | |
6d0bf005 TT |
131 | { |
132 | #ifdef CONFIG_JBD2_DEBUG | |
6d0bf005 TT |
133 | if (jbd2_journal_enable_debug && |
134 | time_after(transaction->t_start, ts)) { | |
135 | ts = jbd2_time_diff(ts, transaction->t_start); | |
136 | spin_lock(&transaction->t_handle_lock); | |
137 | if (ts > transaction->t_max_wait) | |
138 | transaction->t_max_wait = ts; | |
139 | spin_unlock(&transaction->t_handle_lock); | |
140 | } | |
141 | #endif | |
142 | } | |
143 | ||
8f7d89f3 JK |
144 | /* |
145 | * Wait until running transaction passes T_LOCKED state. Also starts the commit | |
146 | * if needed. The function expects running transaction to exist and releases | |
147 | * j_state_lock. | |
148 | */ | |
149 | static void wait_transaction_locked(journal_t *journal) | |
150 | __releases(journal->j_state_lock) | |
151 | { | |
152 | DEFINE_WAIT(wait); | |
153 | int need_to_start; | |
154 | tid_t tid = journal->j_running_transaction->t_tid; | |
155 | ||
156 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, | |
157 | TASK_UNINTERRUPTIBLE); | |
158 | need_to_start = !tid_geq(journal->j_commit_request, tid); | |
159 | read_unlock(&journal->j_state_lock); | |
160 | if (need_to_start) | |
161 | jbd2_log_start_commit(journal, tid); | |
162 | schedule(); | |
163 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
164 | } | |
165 | ||
166 | static void sub_reserved_credits(journal_t *journal, int blocks) | |
167 | { | |
168 | atomic_sub(blocks, &journal->j_reserved_credits); | |
169 | wake_up(&journal->j_wait_reserved); | |
170 | } | |
171 | ||
172 | /* | |
173 | * Wait until we can add credits for handle to the running transaction. Called | |
174 | * with j_state_lock held for reading. Returns 0 if handle joined the running | |
175 | * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and | |
176 | * caller must retry. | |
177 | */ | |
178 | static int add_transaction_credits(journal_t *journal, int blocks, | |
179 | int rsv_blocks) | |
180 | { | |
181 | transaction_t *t = journal->j_running_transaction; | |
182 | int needed; | |
183 | int total = blocks + rsv_blocks; | |
184 | ||
185 | /* | |
186 | * If the current transaction is locked down for commit, wait | |
187 | * for the lock to be released. | |
188 | */ | |
189 | if (t->t_state == T_LOCKED) { | |
190 | wait_transaction_locked(journal); | |
191 | return 1; | |
192 | } | |
193 | ||
194 | /* | |
195 | * If there is not enough space left in the log to write all | |
196 | * potential buffers requested by this operation, we need to | |
197 | * stall pending a log checkpoint to free some more log space. | |
198 | */ | |
199 | needed = atomic_add_return(total, &t->t_outstanding_credits); | |
200 | if (needed > journal->j_max_transaction_buffers) { | |
201 | /* | |
202 | * If the current transaction is already too large, | |
203 | * then start to commit it: we can then go back and | |
204 | * attach this handle to a new transaction. | |
205 | */ | |
206 | atomic_sub(total, &t->t_outstanding_credits); | |
6d3ec14d LC |
207 | |
208 | /* | |
209 | * Is the number of reserved credits in the current transaction too | |
210 | * big to fit this handle? Wait until reserved credits are freed. | |
211 | */ | |
212 | if (atomic_read(&journal->j_reserved_credits) + total > | |
213 | journal->j_max_transaction_buffers) { | |
214 | read_unlock(&journal->j_state_lock); | |
215 | wait_event(journal->j_wait_reserved, | |
216 | atomic_read(&journal->j_reserved_credits) + total <= | |
217 | journal->j_max_transaction_buffers); | |
218 | return 1; | |
219 | } | |
220 | ||
8f7d89f3 JK |
221 | wait_transaction_locked(journal); |
222 | return 1; | |
223 | } | |
224 | ||
225 | /* | |
226 | * The commit code assumes that it can get enough log space | |
227 | * without forcing a checkpoint. This is *critical* for | |
228 | * correctness: a checkpoint of a buffer which is also | |
229 | * associated with a committing transaction creates a deadlock, | |
230 | * so commit simply cannot force through checkpoints. | |
231 | * | |
232 | * We must therefore ensure the necessary space in the journal | |
233 | * *before* starting to dirty potentially checkpointed buffers | |
234 | * in the new transaction. | |
235 | */ | |
236 | if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) { | |
237 | atomic_sub(total, &t->t_outstanding_credits); | |
238 | read_unlock(&journal->j_state_lock); | |
239 | write_lock(&journal->j_state_lock); | |
240 | if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) | |
241 | __jbd2_log_wait_for_space(journal); | |
242 | write_unlock(&journal->j_state_lock); | |
243 | return 1; | |
244 | } | |
245 | ||
246 | /* No reservation? We are done... */ | |
247 | if (!rsv_blocks) | |
248 | return 0; | |
249 | ||
250 | needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits); | |
251 | /* We allow at most half of a transaction to be reserved */ | |
252 | if (needed > journal->j_max_transaction_buffers / 2) { | |
253 | sub_reserved_credits(journal, rsv_blocks); | |
254 | atomic_sub(total, &t->t_outstanding_credits); | |
255 | read_unlock(&journal->j_state_lock); | |
256 | wait_event(journal->j_wait_reserved, | |
257 | atomic_read(&journal->j_reserved_credits) + rsv_blocks | |
258 | <= journal->j_max_transaction_buffers / 2); | |
259 | return 1; | |
260 | } | |
261 | return 0; | |
262 | } | |
263 | ||
470decc6 DK |
264 | /* |
265 | * start_this_handle: Given a handle, deal with any locking or stalling | |
266 | * needed to make sure that there is enough journal space for the handle | |
267 | * to begin. Attach the handle to a transaction and set up the | |
268 | * transaction's buffer credits. | |
269 | */ | |
270 | ||
47def826 | 271 | static int start_this_handle(journal_t *journal, handle_t *handle, |
d2159fb7 | 272 | gfp_t gfp_mask) |
470decc6 | 273 | { |
e4471831 | 274 | transaction_t *transaction, *new_transaction = NULL; |
8f7d89f3 JK |
275 | int blocks = handle->h_buffer_credits; |
276 | int rsv_blocks = 0; | |
28e35e42 | 277 | unsigned long ts = jiffies; |
470decc6 | 278 | |
6d3ec14d LC |
279 | if (handle->h_rsv_handle) |
280 | rsv_blocks = handle->h_rsv_handle->h_buffer_credits; | |
281 | ||
8f7d89f3 | 282 | /* |
6d3ec14d LC |
283 | * Limit the number of reserved credits to 1/2 of maximum transaction |
284 | * size and limit the number of total credits to not exceed maximum | |
285 | * transaction size per operation. | |
8f7d89f3 | 286 | */ |
6d3ec14d LC |
287 | if ((rsv_blocks > journal->j_max_transaction_buffers / 2) || |
288 | (rsv_blocks + blocks > journal->j_max_transaction_buffers)) { | |
289 | printk(KERN_ERR "JBD2: %s wants too many credits " | |
290 | "credits:%d rsv_credits:%d max:%d\n", | |
291 | current->comm, blocks, rsv_blocks, | |
292 | journal->j_max_transaction_buffers); | |
293 | WARN_ON(1); | |
47def826 | 294 | return -ENOSPC; |
470decc6 DK |
295 | } |
296 | ||
297 | alloc_transaction: | |
298 | if (!journal->j_running_transaction) { | |
6ccaf3e2 MH |
299 | /* |
300 | * If __GFP_FS is not present, then we may be being called from | |
301 | * inside the fs writeback layer, so we MUST NOT fail. | |
302 | */ | |
303 | if ((gfp_mask & __GFP_FS) == 0) | |
304 | gfp_mask |= __GFP_NOFAIL; | |
b2f4edb3 WG |
305 | new_transaction = kmem_cache_zalloc(transaction_cache, |
306 | gfp_mask); | |
6ccaf3e2 | 307 | if (!new_transaction) |
47def826 | 308 | return -ENOMEM; |
470decc6 DK |
309 | } |
310 | ||
311 | jbd_debug(3, "New handle %p going live.\n", handle); | |
312 | ||
470decc6 DK |
313 | /* |
314 | * We need to hold j_state_lock until t_updates has been incremented, | |
315 | * for proper journal barrier handling | |
316 | */ | |
a931da6a TT |
317 | repeat: |
318 | read_lock(&journal->j_state_lock); | |
5c2178e7 | 319 | BUG_ON(journal->j_flags & JBD2_UNMOUNT); |
470decc6 | 320 | if (is_journal_aborted(journal) || |
f7f4bccb | 321 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
a931da6a | 322 | read_unlock(&journal->j_state_lock); |
0c2022ec | 323 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 324 | return -EROFS; |
470decc6 DK |
325 | } |
326 | ||
8f7d89f3 JK |
327 | /* |
328 | * Wait on the journal's transaction barrier if necessary. Specifically | |
329 | * we allow reserved handles to proceed because otherwise commit could | |
330 | * deadlock on page writeback not being able to complete. | |
331 | */ | |
332 | if (!handle->h_reserved && journal->j_barrier_count) { | |
a931da6a | 333 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
334 | wait_event(journal->j_wait_transaction_locked, |
335 | journal->j_barrier_count == 0); | |
336 | goto repeat; | |
337 | } | |
338 | ||
339 | if (!journal->j_running_transaction) { | |
a931da6a TT |
340 | read_unlock(&journal->j_state_lock); |
341 | if (!new_transaction) | |
470decc6 | 342 | goto alloc_transaction; |
a931da6a | 343 | write_lock(&journal->j_state_lock); |
d7961c7f | 344 | if (!journal->j_running_transaction && |
8f7d89f3 | 345 | (handle->h_reserved || !journal->j_barrier_count)) { |
a931da6a TT |
346 | jbd2_get_transaction(journal, new_transaction); |
347 | new_transaction = NULL; | |
470decc6 | 348 | } |
a931da6a TT |
349 | write_unlock(&journal->j_state_lock); |
350 | goto repeat; | |
470decc6 DK |
351 | } |
352 | ||
353 | transaction = journal->j_running_transaction; | |
354 | ||
8f7d89f3 JK |
355 | if (!handle->h_reserved) { |
356 | /* We may have dropped j_state_lock - restart in that case */ | |
357 | if (add_transaction_credits(journal, blocks, rsv_blocks)) | |
358 | goto repeat; | |
359 | } else { | |
470decc6 | 360 | /* |
8f7d89f3 JK |
361 | * We have handle reserved so we are allowed to join T_LOCKED |
362 | * transaction and we don't have to check for transaction size | |
363 | * and journal space. | |
470decc6 | 364 | */ |
8f7d89f3 JK |
365 | sub_reserved_credits(journal, blocks); |
366 | handle->h_reserved = 0; | |
470decc6 DK |
367 | } |
368 | ||
369 | /* OK, account for the buffers that this operation expects to | |
8dd42046 | 370 | * use and add the handle to the running transaction. |
8dd42046 | 371 | */ |
28e35e42 | 372 | update_t_max_wait(transaction, ts); |
470decc6 | 373 | handle->h_transaction = transaction; |
8f7d89f3 | 374 | handle->h_requested_credits = blocks; |
343d9c28 | 375 | handle->h_start_jiffies = jiffies; |
a51dca9c | 376 | atomic_inc(&transaction->t_updates); |
8dd42046 | 377 | atomic_inc(&transaction->t_handle_count); |
8f7d89f3 JK |
378 | jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n", |
379 | handle, blocks, | |
a51dca9c | 380 | atomic_read(&transaction->t_outstanding_credits), |
76c39904 | 381 | jbd2_log_space_left(journal)); |
a931da6a | 382 | read_unlock(&journal->j_state_lock); |
41a5b913 | 383 | current->journal_info = handle; |
9599b0e5 JK |
384 | |
385 | lock_map_acquire(&handle->h_lockdep_map); | |
0c2022ec | 386 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 387 | return 0; |
470decc6 DK |
388 | } |
389 | ||
7b751066 MC |
390 | static struct lock_class_key jbd2_handle_key; |
391 | ||
470decc6 DK |
392 | /* Allocate a new handle. This should probably be in a slab... */ |
393 | static handle_t *new_handle(int nblocks) | |
394 | { | |
af1e76d6 | 395 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
396 | if (!handle) |
397 | return NULL; | |
470decc6 DK |
398 | handle->h_buffer_credits = nblocks; |
399 | handle->h_ref = 1; | |
400 | ||
7b751066 MC |
401 | lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle", |
402 | &jbd2_handle_key, 0); | |
403 | ||
470decc6 DK |
404 | return handle; |
405 | } | |
406 | ||
407 | /** | |
f7f4bccb | 408 | * handle_t *jbd2_journal_start() - Obtain a new handle. |
470decc6 DK |
409 | * @journal: Journal to start transaction on. |
410 | * @nblocks: number of block buffer we might modify | |
411 | * | |
412 | * We make sure that the transaction can guarantee at least nblocks of | |
413 | * modified buffers in the log. We block until the log can guarantee | |
8f7d89f3 JK |
414 | * that much space. Additionally, if rsv_blocks > 0, we also create another |
415 | * handle with rsv_blocks reserved blocks in the journal. This handle is | |
416 | * is stored in h_rsv_handle. It is not attached to any particular transaction | |
417 | * and thus doesn't block transaction commit. If the caller uses this reserved | |
418 | * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop() | |
419 | * on the parent handle will dispose the reserved one. Reserved handle has to | |
420 | * be converted to a normal handle using jbd2_journal_start_reserved() before | |
421 | * it can be used. | |
470decc6 | 422 | * |
c867516d EG |
423 | * Return a pointer to a newly allocated handle, or an ERR_PTR() value |
424 | * on failure. | |
470decc6 | 425 | */ |
8f7d89f3 JK |
426 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks, |
427 | gfp_t gfp_mask, unsigned int type, | |
428 | unsigned int line_no) | |
470decc6 DK |
429 | { |
430 | handle_t *handle = journal_current_handle(); | |
431 | int err; | |
432 | ||
433 | if (!journal) | |
434 | return ERR_PTR(-EROFS); | |
435 | ||
436 | if (handle) { | |
437 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
438 | handle->h_ref++; | |
439 | return handle; | |
440 | } | |
441 | ||
442 | handle = new_handle(nblocks); | |
443 | if (!handle) | |
444 | return ERR_PTR(-ENOMEM); | |
8f7d89f3 JK |
445 | if (rsv_blocks) { |
446 | handle_t *rsv_handle; | |
447 | ||
448 | rsv_handle = new_handle(rsv_blocks); | |
449 | if (!rsv_handle) { | |
450 | jbd2_free_handle(handle); | |
451 | return ERR_PTR(-ENOMEM); | |
452 | } | |
453 | rsv_handle->h_reserved = 1; | |
454 | rsv_handle->h_journal = journal; | |
455 | handle->h_rsv_handle = rsv_handle; | |
456 | } | |
470decc6 | 457 | |
47def826 | 458 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 459 | if (err < 0) { |
8f7d89f3 JK |
460 | if (handle->h_rsv_handle) |
461 | jbd2_free_handle(handle->h_rsv_handle); | |
af1e76d6 | 462 | jbd2_free_handle(handle); |
df05c1b8 | 463 | return ERR_PTR(err); |
470decc6 | 464 | } |
343d9c28 TT |
465 | handle->h_type = type; |
466 | handle->h_line_no = line_no; | |
467 | trace_jbd2_handle_start(journal->j_fs_dev->bd_dev, | |
468 | handle->h_transaction->t_tid, type, | |
469 | line_no, nblocks); | |
470decc6 DK |
470 | return handle; |
471 | } | |
47def826 TT |
472 | EXPORT_SYMBOL(jbd2__journal_start); |
473 | ||
474 | ||
475 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |
476 | { | |
8f7d89f3 | 477 | return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0); |
47def826 TT |
478 | } |
479 | EXPORT_SYMBOL(jbd2_journal_start); | |
480 | ||
8f7d89f3 JK |
481 | void jbd2_journal_free_reserved(handle_t *handle) |
482 | { | |
483 | journal_t *journal = handle->h_journal; | |
484 | ||
485 | WARN_ON(!handle->h_reserved); | |
486 | sub_reserved_credits(journal, handle->h_buffer_credits); | |
487 | jbd2_free_handle(handle); | |
488 | } | |
489 | EXPORT_SYMBOL(jbd2_journal_free_reserved); | |
490 | ||
491 | /** | |
492 | * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle | |
493 | * @handle: handle to start | |
494 | * | |
495 | * Start handle that has been previously reserved with jbd2_journal_reserve(). | |
496 | * This attaches @handle to the running transaction (or creates one if there's | |
497 | * not transaction running). Unlike jbd2_journal_start() this function cannot | |
498 | * block on journal commit, checkpointing, or similar stuff. It can block on | |
499 | * memory allocation or frozen journal though. | |
500 | * | |
501 | * Return 0 on success, non-zero on error - handle is freed in that case. | |
502 | */ | |
503 | int jbd2_journal_start_reserved(handle_t *handle, unsigned int type, | |
504 | unsigned int line_no) | |
505 | { | |
506 | journal_t *journal = handle->h_journal; | |
507 | int ret = -EIO; | |
508 | ||
509 | if (WARN_ON(!handle->h_reserved)) { | |
510 | /* Someone passed in normal handle? Just stop it. */ | |
511 | jbd2_journal_stop(handle); | |
512 | return ret; | |
513 | } | |
514 | /* | |
515 | * Usefulness of mixing of reserved and unreserved handles is | |
516 | * questionable. So far nobody seems to need it so just error out. | |
517 | */ | |
518 | if (WARN_ON(current->journal_info)) { | |
519 | jbd2_journal_free_reserved(handle); | |
520 | return ret; | |
521 | } | |
522 | ||
523 | handle->h_journal = NULL; | |
8f7d89f3 JK |
524 | /* |
525 | * GFP_NOFS is here because callers are likely from writeback or | |
526 | * similarly constrained call sites | |
527 | */ | |
528 | ret = start_this_handle(journal, handle, GFP_NOFS); | |
92e3b405 | 529 | if (ret < 0) { |
8f7d89f3 | 530 | jbd2_journal_free_reserved(handle); |
92e3b405 DC |
531 | return ret; |
532 | } | |
8f7d89f3 JK |
533 | handle->h_type = type; |
534 | handle->h_line_no = line_no; | |
92e3b405 | 535 | return 0; |
8f7d89f3 JK |
536 | } |
537 | EXPORT_SYMBOL(jbd2_journal_start_reserved); | |
470decc6 DK |
538 | |
539 | /** | |
f7f4bccb | 540 | * int jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
541 | * @handle: handle to 'extend' |
542 | * @nblocks: nr blocks to try to extend by. | |
543 | * | |
544 | * Some transactions, such as large extends and truncates, can be done | |
545 | * atomically all at once or in several stages. The operation requests | |
546 | * a credit for a number of buffer modications in advance, but can | |
547 | * extend its credit if it needs more. | |
548 | * | |
f7f4bccb | 549 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
550 | * It does not guarantee that allocation - this is a best-effort only. |
551 | * The calling process MUST be able to deal cleanly with a failure to | |
552 | * extend here. | |
553 | * | |
554 | * Return 0 on success, non-zero on failure. | |
555 | * | |
556 | * return code < 0 implies an error | |
557 | * return code > 0 implies normal transaction-full status. | |
558 | */ | |
f7f4bccb | 559 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
470decc6 DK |
560 | { |
561 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 562 | journal_t *journal; |
470decc6 DK |
563 | int result; |
564 | int wanted; | |
565 | ||
470decc6 | 566 | if (is_handle_aborted(handle)) |
41a5b913 TT |
567 | return -EROFS; |
568 | journal = transaction->t_journal; | |
470decc6 DK |
569 | |
570 | result = 1; | |
571 | ||
a931da6a | 572 | read_lock(&journal->j_state_lock); |
470decc6 DK |
573 | |
574 | /* Don't extend a locked-down transaction! */ | |
41a5b913 | 575 | if (transaction->t_state != T_RUNNING) { |
470decc6 DK |
576 | jbd_debug(3, "denied handle %p %d blocks: " |
577 | "transaction not running\n", handle, nblocks); | |
578 | goto error_out; | |
579 | } | |
580 | ||
581 | spin_lock(&transaction->t_handle_lock); | |
fe1e8db5 JK |
582 | wanted = atomic_add_return(nblocks, |
583 | &transaction->t_outstanding_credits); | |
470decc6 DK |
584 | |
585 | if (wanted > journal->j_max_transaction_buffers) { | |
586 | jbd_debug(3, "denied handle %p %d blocks: " | |
587 | "transaction too large\n", handle, nblocks); | |
fe1e8db5 | 588 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
589 | goto unlock; |
590 | } | |
591 | ||
76c39904 JK |
592 | if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) > |
593 | jbd2_log_space_left(journal)) { | |
470decc6 DK |
594 | jbd_debug(3, "denied handle %p %d blocks: " |
595 | "insufficient log space\n", handle, nblocks); | |
fe1e8db5 | 596 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
597 | goto unlock; |
598 | } | |
599 | ||
343d9c28 | 600 | trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev, |
41a5b913 | 601 | transaction->t_tid, |
343d9c28 TT |
602 | handle->h_type, handle->h_line_no, |
603 | handle->h_buffer_credits, | |
604 | nblocks); | |
605 | ||
470decc6 | 606 | handle->h_buffer_credits += nblocks; |
343d9c28 | 607 | handle->h_requested_credits += nblocks; |
470decc6 DK |
608 | result = 0; |
609 | ||
610 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | |
611 | unlock: | |
612 | spin_unlock(&transaction->t_handle_lock); | |
613 | error_out: | |
a931da6a | 614 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
615 | return result; |
616 | } | |
617 | ||
618 | ||
619 | /** | |
f7f4bccb | 620 | * int jbd2_journal_restart() - restart a handle . |
470decc6 DK |
621 | * @handle: handle to restart |
622 | * @nblocks: nr credits requested | |
623 | * | |
624 | * Restart a handle for a multi-transaction filesystem | |
625 | * operation. | |
626 | * | |
f7f4bccb MC |
627 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
628 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 DK |
629 | * handle's transaction so far and reattach the handle to a new |
630 | * transaction capabable of guaranteeing the requested number of | |
8f7d89f3 JK |
631 | * credits. We preserve reserved handle if there's any attached to the |
632 | * passed in handle. | |
470decc6 | 633 | */ |
d2159fb7 | 634 | int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask) |
470decc6 DK |
635 | { |
636 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 637 | journal_t *journal; |
e4471831 TT |
638 | tid_t tid; |
639 | int need_to_start, ret; | |
470decc6 DK |
640 | |
641 | /* If we've had an abort of any type, don't even think about | |
642 | * actually doing the restart! */ | |
643 | if (is_handle_aborted(handle)) | |
644 | return 0; | |
41a5b913 | 645 | journal = transaction->t_journal; |
470decc6 DK |
646 | |
647 | /* | |
648 | * First unlink the handle from its current transaction, and start the | |
649 | * commit on that. | |
650 | */ | |
a51dca9c | 651 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
652 | J_ASSERT(journal_current_handle() == handle); |
653 | ||
a931da6a | 654 | read_lock(&journal->j_state_lock); |
470decc6 | 655 | spin_lock(&transaction->t_handle_lock); |
a51dca9c TT |
656 | atomic_sub(handle->h_buffer_credits, |
657 | &transaction->t_outstanding_credits); | |
8f7d89f3 JK |
658 | if (handle->h_rsv_handle) { |
659 | sub_reserved_credits(journal, | |
660 | handle->h_rsv_handle->h_buffer_credits); | |
661 | } | |
a51dca9c | 662 | if (atomic_dec_and_test(&transaction->t_updates)) |
470decc6 | 663 | wake_up(&journal->j_wait_updates); |
39c04153 | 664 | tid = transaction->t_tid; |
470decc6 | 665 | spin_unlock(&transaction->t_handle_lock); |
41a5b913 TT |
666 | handle->h_transaction = NULL; |
667 | current->journal_info = NULL; | |
470decc6 DK |
668 | |
669 | jbd_debug(2, "restarting handle %p\n", handle); | |
e4471831 | 670 | need_to_start = !tid_geq(journal->j_commit_request, tid); |
a931da6a | 671 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
672 | if (need_to_start) |
673 | jbd2_log_start_commit(journal, tid); | |
470decc6 | 674 | |
9599b0e5 | 675 | lock_map_release(&handle->h_lockdep_map); |
470decc6 | 676 | handle->h_buffer_credits = nblocks; |
47def826 | 677 | ret = start_this_handle(journal, handle, gfp_mask); |
470decc6 DK |
678 | return ret; |
679 | } | |
47def826 | 680 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
681 | |
682 | ||
47def826 TT |
683 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
684 | { | |
685 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | |
686 | } | |
687 | EXPORT_SYMBOL(jbd2_journal_restart); | |
688 | ||
470decc6 | 689 | /** |
f7f4bccb | 690 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
691 | * @journal: Journal to establish a barrier on. |
692 | * | |
693 | * This locks out any further updates from being started, and blocks | |
694 | * until all existing updates have completed, returning only once the | |
695 | * journal is in a quiescent state with no updates running. | |
696 | * | |
697 | * The journal lock should not be held on entry. | |
698 | */ | |
f7f4bccb | 699 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 DK |
700 | { |
701 | DEFINE_WAIT(wait); | |
702 | ||
a931da6a | 703 | write_lock(&journal->j_state_lock); |
470decc6 DK |
704 | ++journal->j_barrier_count; |
705 | ||
8f7d89f3 JK |
706 | /* Wait until there are no reserved handles */ |
707 | if (atomic_read(&journal->j_reserved_credits)) { | |
708 | write_unlock(&journal->j_state_lock); | |
709 | wait_event(journal->j_wait_reserved, | |
710 | atomic_read(&journal->j_reserved_credits) == 0); | |
711 | write_lock(&journal->j_state_lock); | |
712 | } | |
713 | ||
470decc6 DK |
714 | /* Wait until there are no running updates */ |
715 | while (1) { | |
716 | transaction_t *transaction = journal->j_running_transaction; | |
717 | ||
718 | if (!transaction) | |
719 | break; | |
720 | ||
721 | spin_lock(&transaction->t_handle_lock); | |
9837d8e9 JK |
722 | prepare_to_wait(&journal->j_wait_updates, &wait, |
723 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 724 | if (!atomic_read(&transaction->t_updates)) { |
470decc6 | 725 | spin_unlock(&transaction->t_handle_lock); |
9837d8e9 | 726 | finish_wait(&journal->j_wait_updates, &wait); |
470decc6 DK |
727 | break; |
728 | } | |
470decc6 | 729 | spin_unlock(&transaction->t_handle_lock); |
a931da6a | 730 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
731 | schedule(); |
732 | finish_wait(&journal->j_wait_updates, &wait); | |
a931da6a | 733 | write_lock(&journal->j_state_lock); |
470decc6 | 734 | } |
a931da6a | 735 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
736 | |
737 | /* | |
738 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 739 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
740 | * to make sure that we serialise special journal-locked operations |
741 | * too. | |
742 | */ | |
743 | mutex_lock(&journal->j_barrier); | |
744 | } | |
745 | ||
746 | /** | |
f7f4bccb | 747 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
470decc6 DK |
748 | * @journal: Journal to release the barrier on. |
749 | * | |
f7f4bccb | 750 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
751 | * |
752 | * Should be called without the journal lock held. | |
753 | */ | |
f7f4bccb | 754 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
755 | { |
756 | J_ASSERT(journal->j_barrier_count != 0); | |
757 | ||
758 | mutex_unlock(&journal->j_barrier); | |
a931da6a | 759 | write_lock(&journal->j_state_lock); |
470decc6 | 760 | --journal->j_barrier_count; |
a931da6a | 761 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
762 | wake_up(&journal->j_wait_transaction_locked); |
763 | } | |
764 | ||
f91d1d04 | 765 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 766 | { |
f91d1d04 | 767 | char b[BDEVNAME_SIZE]; |
470decc6 | 768 | |
f91d1d04 | 769 | printk(KERN_WARNING |
f2a44523 | 770 | "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). " |
f91d1d04 JK |
771 | "There's a risk of filesystem corruption in case of system " |
772 | "crash.\n", | |
773 | bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr); | |
470decc6 DK |
774 | } |
775 | ||
ee57aba1 JK |
776 | /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */ |
777 | static void jbd2_freeze_jh_data(struct journal_head *jh) | |
778 | { | |
779 | struct page *page; | |
780 | int offset; | |
781 | char *source; | |
782 | struct buffer_head *bh = jh2bh(jh); | |
783 | ||
784 | J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n"); | |
785 | page = bh->b_page; | |
786 | offset = offset_in_page(bh->b_data); | |
787 | source = kmap_atomic(page); | |
788 | /* Fire data frozen trigger just before we copy the data */ | |
789 | jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers); | |
790 | memcpy(jh->b_frozen_data, source + offset, bh->b_size); | |
791 | kunmap_atomic(source); | |
792 | ||
793 | /* | |
794 | * Now that the frozen data is saved off, we need to store any matching | |
795 | * triggers. | |
796 | */ | |
797 | jh->b_frozen_triggers = jh->b_triggers; | |
798 | } | |
799 | ||
470decc6 DK |
800 | /* |
801 | * If the buffer is already part of the current transaction, then there | |
802 | * is nothing we need to do. If it is already part of a prior | |
803 | * transaction which we are still committing to disk, then we need to | |
804 | * make sure that we do not overwrite the old copy: we do copy-out to | |
805 | * preserve the copy going to disk. We also account the buffer against | |
806 | * the handle's metadata buffer credits (unless the buffer is already | |
807 | * part of the transaction, that is). | |
808 | * | |
809 | */ | |
810 | static int | |
811 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
812 | int force_copy) | |
813 | { | |
814 | struct buffer_head *bh; | |
41a5b913 | 815 | transaction_t *transaction = handle->h_transaction; |
470decc6 DK |
816 | journal_t *journal; |
817 | int error; | |
818 | char *frozen_buffer = NULL; | |
f783f091 | 819 | unsigned long start_lock, time_lock; |
470decc6 DK |
820 | |
821 | if (is_handle_aborted(handle)) | |
822 | return -EROFS; | |
470decc6 DK |
823 | journal = transaction->t_journal; |
824 | ||
cfef2c6a | 825 | jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy); |
470decc6 DK |
826 | |
827 | JBUFFER_TRACE(jh, "entry"); | |
828 | repeat: | |
829 | bh = jh2bh(jh); | |
830 | ||
831 | /* @@@ Need to check for errors here at some point. */ | |
832 | ||
f783f091 | 833 | start_lock = jiffies; |
470decc6 DK |
834 | lock_buffer(bh); |
835 | jbd_lock_bh_state(bh); | |
836 | ||
f783f091 TT |
837 | /* If it takes too long to lock the buffer, trace it */ |
838 | time_lock = jbd2_time_diff(start_lock, jiffies); | |
839 | if (time_lock > HZ/10) | |
840 | trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev, | |
841 | jiffies_to_msecs(time_lock)); | |
842 | ||
470decc6 DK |
843 | /* We now hold the buffer lock so it is safe to query the buffer |
844 | * state. Is the buffer dirty? | |
845 | * | |
846 | * If so, there are two possibilities. The buffer may be | |
847 | * non-journaled, and undergoing a quite legitimate writeback. | |
848 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
849 | * in that state (the buffers should be marked JBD_Dirty | |
850 | * instead.) So either the IO is being done under our own | |
851 | * control and this is a bug, or it's a third party IO such as | |
852 | * dump(8) (which may leave the buffer scheduled for read --- | |
853 | * ie. locked but not dirty) or tune2fs (which may actually have | |
854 | * the buffer dirtied, ugh.) */ | |
855 | ||
856 | if (buffer_dirty(bh)) { | |
857 | /* | |
858 | * First question: is this buffer already part of the current | |
859 | * transaction or the existing committing transaction? | |
860 | */ | |
861 | if (jh->b_transaction) { | |
862 | J_ASSERT_JH(jh, | |
863 | jh->b_transaction == transaction || | |
864 | jh->b_transaction == | |
865 | journal->j_committing_transaction); | |
866 | if (jh->b_next_transaction) | |
867 | J_ASSERT_JH(jh, jh->b_next_transaction == | |
868 | transaction); | |
f91d1d04 | 869 | warn_dirty_buffer(bh); |
470decc6 DK |
870 | } |
871 | /* | |
872 | * In any case we need to clean the dirty flag and we must | |
873 | * do it under the buffer lock to be sure we don't race | |
874 | * with running write-out. | |
875 | */ | |
f91d1d04 JK |
876 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
877 | clear_buffer_dirty(bh); | |
878 | set_buffer_jbddirty(bh); | |
470decc6 DK |
879 | } |
880 | ||
881 | unlock_buffer(bh); | |
882 | ||
883 | error = -EROFS; | |
884 | if (is_handle_aborted(handle)) { | |
885 | jbd_unlock_bh_state(bh); | |
886 | goto out; | |
887 | } | |
888 | error = 0; | |
889 | ||
890 | /* | |
891 | * The buffer is already part of this transaction if b_transaction or | |
892 | * b_next_transaction points to it | |
893 | */ | |
894 | if (jh->b_transaction == transaction || | |
895 | jh->b_next_transaction == transaction) | |
896 | goto done; | |
897 | ||
9fc7c63a JB |
898 | /* |
899 | * this is the first time this transaction is touching this buffer, | |
900 | * reset the modified flag | |
901 | */ | |
902 | jh->b_modified = 0; | |
903 | ||
8b00f400 JK |
904 | /* |
905 | * If the buffer is not journaled right now, we need to make sure it | |
906 | * doesn't get written to disk before the caller actually commits the | |
907 | * new data | |
908 | */ | |
909 | if (!jh->b_transaction) { | |
910 | JBUFFER_TRACE(jh, "no transaction"); | |
911 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
912 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); | |
de92c8ca JK |
913 | /* |
914 | * Make sure all stores to jh (b_modified, b_frozen_data) are | |
915 | * visible before attaching it to the running transaction. | |
916 | * Paired with barrier in jbd2_write_access_granted() | |
917 | */ | |
918 | smp_wmb(); | |
8b00f400 JK |
919 | spin_lock(&journal->j_list_lock); |
920 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); | |
921 | spin_unlock(&journal->j_list_lock); | |
922 | goto done; | |
923 | } | |
470decc6 DK |
924 | /* |
925 | * If there is already a copy-out version of this buffer, then we don't | |
926 | * need to make another one | |
927 | */ | |
928 | if (jh->b_frozen_data) { | |
929 | JBUFFER_TRACE(jh, "has frozen data"); | |
930 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
de92c8ca | 931 | goto attach_next; |
470decc6 DK |
932 | } |
933 | ||
8b00f400 JK |
934 | JBUFFER_TRACE(jh, "owned by older transaction"); |
935 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
936 | J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction); | |
470decc6 | 937 | |
8b00f400 JK |
938 | /* |
939 | * There is one case we have to be very careful about. If the | |
940 | * committing transaction is currently writing this buffer out to disk | |
941 | * and has NOT made a copy-out, then we cannot modify the buffer | |
942 | * contents at all right now. The essence of copy-out is that it is | |
943 | * the extra copy, not the primary copy, which gets journaled. If the | |
944 | * primary copy is already going to disk then we cannot do copy-out | |
945 | * here. | |
946 | */ | |
947 | if (buffer_shadow(bh)) { | |
948 | JBUFFER_TRACE(jh, "on shadow: sleep"); | |
949 | jbd_unlock_bh_state(bh); | |
950 | wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE); | |
951 | goto repeat; | |
952 | } | |
470decc6 | 953 | |
8b00f400 JK |
954 | /* |
955 | * Only do the copy if the currently-owning transaction still needs it. | |
956 | * If buffer isn't on BJ_Metadata list, the committing transaction is | |
957 | * past that stage (here we use the fact that BH_Shadow is set under | |
958 | * bh_state lock together with refiling to BJ_Shadow list and at this | |
959 | * point we know the buffer doesn't have BH_Shadow set). | |
960 | * | |
961 | * Subtle point, though: if this is a get_undo_access, then we will be | |
962 | * relying on the frozen_data to contain the new value of the | |
963 | * committed_data record after the transaction, so we HAVE to force the | |
964 | * frozen_data copy in that case. | |
965 | */ | |
966 | if (jh->b_jlist == BJ_Metadata || force_copy) { | |
967 | JBUFFER_TRACE(jh, "generate frozen data"); | |
968 | if (!frozen_buffer) { | |
969 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
470decc6 | 970 | jbd_unlock_bh_state(bh); |
8b00f400 | 971 | frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS); |
470decc6 | 972 | if (!frozen_buffer) { |
8b00f400 JK |
973 | printk(KERN_ERR "%s: OOM for frozen_buffer\n", |
974 | __func__); | |
975 | JBUFFER_TRACE(jh, "oom!"); | |
976 | error = -ENOMEM; | |
977 | goto out; | |
470decc6 | 978 | } |
8b00f400 | 979 | goto repeat; |
470decc6 | 980 | } |
8b00f400 JK |
981 | jh->b_frozen_data = frozen_buffer; |
982 | frozen_buffer = NULL; | |
983 | jbd2_freeze_jh_data(jh); | |
470decc6 | 984 | } |
de92c8ca JK |
985 | attach_next: |
986 | /* | |
987 | * Make sure all stores to jh (b_modified, b_frozen_data) are visible | |
988 | * before attaching it to the running transaction. Paired with barrier | |
989 | * in jbd2_write_access_granted() | |
990 | */ | |
991 | smp_wmb(); | |
8b00f400 | 992 | jh->b_next_transaction = transaction; |
470decc6 DK |
993 | |
994 | done: | |
470decc6 DK |
995 | jbd_unlock_bh_state(bh); |
996 | ||
997 | /* | |
998 | * If we are about to journal a buffer, then any revoke pending on it is | |
999 | * no longer valid | |
1000 | */ | |
f7f4bccb | 1001 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
1002 | |
1003 | out: | |
1004 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 1005 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
1006 | |
1007 | JBUFFER_TRACE(jh, "exit"); | |
1008 | return error; | |
1009 | } | |
1010 | ||
de92c8ca JK |
1011 | /* Fast check whether buffer is already attached to the required transaction */ |
1012 | static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh) | |
1013 | { | |
1014 | struct journal_head *jh; | |
1015 | bool ret = false; | |
1016 | ||
1017 | /* Dirty buffers require special handling... */ | |
1018 | if (buffer_dirty(bh)) | |
1019 | return false; | |
1020 | ||
1021 | /* | |
1022 | * RCU protects us from dereferencing freed pages. So the checks we do | |
1023 | * are guaranteed not to oops. However the jh slab object can get freed | |
1024 | * & reallocated while we work with it. So we have to be careful. When | |
1025 | * we see jh attached to the running transaction, we know it must stay | |
1026 | * so until the transaction is committed. Thus jh won't be freed and | |
1027 | * will be attached to the same bh while we run. However it can | |
1028 | * happen jh gets freed, reallocated, and attached to the transaction | |
1029 | * just after we get pointer to it from bh. So we have to be careful | |
1030 | * and recheck jh still belongs to our bh before we return success. | |
1031 | */ | |
1032 | rcu_read_lock(); | |
1033 | if (!buffer_jbd(bh)) | |
1034 | goto out; | |
1035 | /* This should be bh2jh() but that doesn't work with inline functions */ | |
1036 | jh = READ_ONCE(bh->b_private); | |
1037 | if (!jh) | |
1038 | goto out; | |
1039 | if (jh->b_transaction != handle->h_transaction && | |
1040 | jh->b_next_transaction != handle->h_transaction) | |
1041 | goto out; | |
1042 | /* | |
1043 | * There are two reasons for the barrier here: | |
1044 | * 1) Make sure to fetch b_bh after we did previous checks so that we | |
1045 | * detect when jh went through free, realloc, attach to transaction | |
1046 | * while we were checking. Paired with implicit barrier in that path. | |
1047 | * 2) So that access to bh done after jbd2_write_access_granted() | |
1048 | * doesn't get reordered and see inconsistent state of concurrent | |
1049 | * do_get_write_access(). | |
1050 | */ | |
1051 | smp_mb(); | |
1052 | if (unlikely(jh->b_bh != bh)) | |
1053 | goto out; | |
1054 | ret = true; | |
1055 | out: | |
1056 | rcu_read_unlock(); | |
1057 | return ret; | |
1058 | } | |
1059 | ||
470decc6 | 1060 | /** |
f7f4bccb | 1061 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
470decc6 DK |
1062 | * @handle: transaction to add buffer modifications to |
1063 | * @bh: bh to be used for metadata writes | |
470decc6 DK |
1064 | * |
1065 | * Returns an error code or 0 on success. | |
1066 | * | |
1067 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
1068 | * because we're write()ing a buffer which is also part of a shared mapping. | |
1069 | */ | |
1070 | ||
f7f4bccb | 1071 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 1072 | { |
de92c8ca | 1073 | struct journal_head *jh; |
470decc6 DK |
1074 | int rc; |
1075 | ||
de92c8ca JK |
1076 | if (jbd2_write_access_granted(handle, bh)) |
1077 | return 0; | |
1078 | ||
1079 | jh = jbd2_journal_add_journal_head(bh); | |
470decc6 DK |
1080 | /* We do not want to get caught playing with fields which the |
1081 | * log thread also manipulates. Make sure that the buffer | |
1082 | * completes any outstanding IO before proceeding. */ | |
1083 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 1084 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1085 | return rc; |
1086 | } | |
1087 | ||
1088 | ||
1089 | /* | |
1090 | * When the user wants to journal a newly created buffer_head | |
1091 | * (ie. getblk() returned a new buffer and we are going to populate it | |
1092 | * manually rather than reading off disk), then we need to keep the | |
1093 | * buffer_head locked until it has been completely filled with new | |
1094 | * data. In this case, we should be able to make the assertion that | |
1095 | * the bh is not already part of an existing transaction. | |
1096 | * | |
1097 | * The buffer should already be locked by the caller by this point. | |
1098 | * There is no lock ranking violation: it was a newly created, | |
1099 | * unlocked buffer beforehand. */ | |
1100 | ||
1101 | /** | |
f7f4bccb | 1102 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
1103 | * @handle: transaction to new buffer to |
1104 | * @bh: new buffer. | |
1105 | * | |
1106 | * Call this if you create a new bh. | |
1107 | */ | |
f7f4bccb | 1108 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1109 | { |
1110 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1111 | journal_t *journal; |
f7f4bccb | 1112 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
1113 | int err; |
1114 | ||
1115 | jbd_debug(5, "journal_head %p\n", jh); | |
1116 | err = -EROFS; | |
1117 | if (is_handle_aborted(handle)) | |
1118 | goto out; | |
41a5b913 | 1119 | journal = transaction->t_journal; |
470decc6 DK |
1120 | err = 0; |
1121 | ||
1122 | JBUFFER_TRACE(jh, "entry"); | |
1123 | /* | |
1124 | * The buffer may already belong to this transaction due to pre-zeroing | |
1125 | * in the filesystem's new_block code. It may also be on the previous, | |
1126 | * committing transaction's lists, but it HAS to be in Forget state in | |
1127 | * that case: the transaction must have deleted the buffer for it to be | |
1128 | * reused here. | |
1129 | */ | |
1130 | jbd_lock_bh_state(bh); | |
470decc6 DK |
1131 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || |
1132 | jh->b_transaction == NULL || | |
1133 | (jh->b_transaction == journal->j_committing_transaction && | |
1134 | jh->b_jlist == BJ_Forget))); | |
1135 | ||
1136 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
1137 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
1138 | ||
1139 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
1140 | /* |
1141 | * Previous jbd2_journal_forget() could have left the buffer | |
1142 | * with jbddirty bit set because it was being committed. When | |
1143 | * the commit finished, we've filed the buffer for | |
1144 | * checkpointing and marked it dirty. Now we are reallocating | |
1145 | * the buffer so the transaction freeing it must have | |
1146 | * committed and so it's safe to clear the dirty bit. | |
1147 | */ | |
1148 | clear_buffer_dirty(jh2bh(jh)); | |
9fc7c63a JB |
1149 | /* first access by this transaction */ |
1150 | jh->b_modified = 0; | |
1151 | ||
470decc6 | 1152 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
6e4862a5 | 1153 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 1154 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 | 1155 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
1156 | /* first access by this transaction */ |
1157 | jh->b_modified = 0; | |
1158 | ||
470decc6 | 1159 | JBUFFER_TRACE(jh, "set next transaction"); |
6e4862a5 | 1160 | spin_lock(&journal->j_list_lock); |
470decc6 DK |
1161 | jh->b_next_transaction = transaction; |
1162 | } | |
1163 | spin_unlock(&journal->j_list_lock); | |
1164 | jbd_unlock_bh_state(bh); | |
1165 | ||
1166 | /* | |
1167 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
1168 | * blocks which contain freed but then revoked metadata. We need | |
1169 | * to cancel the revoke in case we end up freeing it yet again | |
1170 | * and the reallocating as data - this would cause a second revoke, | |
1171 | * which hits an assertion error. | |
1172 | */ | |
1173 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb | 1174 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 | 1175 | out: |
3991b400 | 1176 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1177 | return err; |
1178 | } | |
1179 | ||
1180 | /** | |
f7f4bccb | 1181 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
1182 | * non-rewindable consequences |
1183 | * @handle: transaction | |
1184 | * @bh: buffer to undo | |
470decc6 DK |
1185 | * |
1186 | * Sometimes there is a need to distinguish between metadata which has | |
1187 | * been committed to disk and that which has not. The ext3fs code uses | |
1188 | * this for freeing and allocating space, we have to make sure that we | |
1189 | * do not reuse freed space until the deallocation has been committed, | |
1190 | * since if we overwrote that space we would make the delete | |
1191 | * un-rewindable in case of a crash. | |
1192 | * | |
f7f4bccb | 1193 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
1194 | * buffer for parts of non-rewindable operations such as delete |
1195 | * operations on the bitmaps. The journaling code must keep a copy of | |
1196 | * the buffer's contents prior to the undo_access call until such time | |
1197 | * as we know that the buffer has definitely been committed to disk. | |
1198 | * | |
1199 | * We never need to know which transaction the committed data is part | |
1200 | * of, buffers touched here are guaranteed to be dirtied later and so | |
1201 | * will be committed to a new transaction in due course, at which point | |
1202 | * we can discard the old committed data pointer. | |
1203 | * | |
1204 | * Returns error number or 0 on success. | |
1205 | */ | |
f7f4bccb | 1206 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1207 | { |
1208 | int err; | |
de92c8ca | 1209 | struct journal_head *jh; |
470decc6 DK |
1210 | char *committed_data = NULL; |
1211 | ||
1212 | JBUFFER_TRACE(jh, "entry"); | |
de92c8ca JK |
1213 | if (jbd2_write_access_granted(handle, bh)) |
1214 | return 0; | |
470decc6 | 1215 | |
de92c8ca | 1216 | jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
1217 | /* |
1218 | * Do this first --- it can drop the journal lock, so we want to | |
1219 | * make sure that obtaining the committed_data is done | |
1220 | * atomically wrt. completion of any outstanding commits. | |
1221 | */ | |
1222 | err = do_get_write_access(handle, jh, 1); | |
1223 | if (err) | |
1224 | goto out; | |
1225 | ||
1226 | repeat: | |
1227 | if (!jh->b_committed_data) { | |
af1e76d6 | 1228 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS); |
470decc6 | 1229 | if (!committed_data) { |
75685071 | 1230 | printk(KERN_ERR "%s: No memory for committed data\n", |
329d291f | 1231 | __func__); |
470decc6 DK |
1232 | err = -ENOMEM; |
1233 | goto out; | |
1234 | } | |
1235 | } | |
1236 | ||
1237 | jbd_lock_bh_state(bh); | |
1238 | if (!jh->b_committed_data) { | |
1239 | /* Copy out the current buffer contents into the | |
1240 | * preserved, committed copy. */ | |
1241 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
1242 | if (!committed_data) { | |
1243 | jbd_unlock_bh_state(bh); | |
1244 | goto repeat; | |
1245 | } | |
1246 | ||
1247 | jh->b_committed_data = committed_data; | |
1248 | committed_data = NULL; | |
1249 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
1250 | } | |
1251 | jbd_unlock_bh_state(bh); | |
1252 | out: | |
f7f4bccb | 1253 | jbd2_journal_put_journal_head(jh); |
470decc6 | 1254 | if (unlikely(committed_data)) |
af1e76d6 | 1255 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
1256 | return err; |
1257 | } | |
1258 | ||
e06c8227 JB |
1259 | /** |
1260 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout | |
1261 | * @bh: buffer to trigger on | |
1262 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
1263 | * | |
1264 | * Set any triggers on this journal_head. This is always safe, because | |
1265 | * triggers for a committing buffer will be saved off, and triggers for | |
1266 | * a running transaction will match the buffer in that transaction. | |
1267 | * | |
1268 | * Call with NULL to clear the triggers. | |
1269 | */ | |
1270 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
1271 | struct jbd2_buffer_trigger_type *type) | |
1272 | { | |
ad56edad | 1273 | struct journal_head *jh = jbd2_journal_grab_journal_head(bh); |
e06c8227 | 1274 | |
ad56edad JK |
1275 | if (WARN_ON(!jh)) |
1276 | return; | |
e06c8227 | 1277 | jh->b_triggers = type; |
ad56edad | 1278 | jbd2_journal_put_journal_head(jh); |
e06c8227 JB |
1279 | } |
1280 | ||
13ceef09 | 1281 | void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data, |
e06c8227 JB |
1282 | struct jbd2_buffer_trigger_type *triggers) |
1283 | { | |
1284 | struct buffer_head *bh = jh2bh(jh); | |
1285 | ||
13ceef09 | 1286 | if (!triggers || !triggers->t_frozen) |
e06c8227 JB |
1287 | return; |
1288 | ||
13ceef09 | 1289 | triggers->t_frozen(triggers, bh, mapped_data, bh->b_size); |
e06c8227 JB |
1290 | } |
1291 | ||
1292 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1293 | struct jbd2_buffer_trigger_type *triggers) | |
1294 | { | |
1295 | if (!triggers || !triggers->t_abort) | |
1296 | return; | |
1297 | ||
1298 | triggers->t_abort(triggers, jh2bh(jh)); | |
1299 | } | |
1300 | ||
470decc6 | 1301 | /** |
f7f4bccb | 1302 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1303 | * @handle: transaction to add buffer to. |
1304 | * @bh: buffer to mark | |
1305 | * | |
1306 | * mark dirty metadata which needs to be journaled as part of the current | |
1307 | * transaction. | |
1308 | * | |
9ea7a0df TT |
1309 | * The buffer must have previously had jbd2_journal_get_write_access() |
1310 | * called so that it has a valid journal_head attached to the buffer | |
1311 | * head. | |
1312 | * | |
470decc6 DK |
1313 | * The buffer is placed on the transaction's metadata list and is marked |
1314 | * as belonging to the transaction. | |
1315 | * | |
1316 | * Returns error number or 0 on success. | |
1317 | * | |
1318 | * Special care needs to be taken if the buffer already belongs to the | |
1319 | * current committing transaction (in which case we should have frozen | |
1320 | * data present for that commit). In that case, we don't relink the | |
1321 | * buffer: that only gets done when the old transaction finally | |
1322 | * completes its commit. | |
1323 | */ | |
f7f4bccb | 1324 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1325 | { |
1326 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1327 | journal_t *journal; |
ad56edad | 1328 | struct journal_head *jh; |
9ea7a0df | 1329 | int ret = 0; |
470decc6 | 1330 | |
470decc6 | 1331 | if (is_handle_aborted(handle)) |
41a5b913 | 1332 | return -EROFS; |
6e06ae88 | 1333 | if (!buffer_jbd(bh)) { |
9ea7a0df TT |
1334 | ret = -EUCLEAN; |
1335 | goto out; | |
1336 | } | |
6e06ae88 JK |
1337 | /* |
1338 | * We don't grab jh reference here since the buffer must be part | |
1339 | * of the running transaction. | |
1340 | */ | |
1341 | jh = bh2jh(bh); | |
1342 | /* | |
1343 | * This and the following assertions are unreliable since we may see jh | |
1344 | * in inconsistent state unless we grab bh_state lock. But this is | |
1345 | * crucial to catch bugs so let's do a reliable check until the | |
1346 | * lockless handling is fully proven. | |
1347 | */ | |
1348 | if (jh->b_transaction != transaction && | |
1349 | jh->b_next_transaction != transaction) { | |
1350 | jbd_lock_bh_state(bh); | |
1351 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
1352 | jh->b_next_transaction == transaction); | |
1353 | jbd_unlock_bh_state(bh); | |
1354 | } | |
1355 | if (jh->b_modified == 1) { | |
1356 | /* If it's in our transaction it must be in BJ_Metadata list. */ | |
1357 | if (jh->b_transaction == transaction && | |
1358 | jh->b_jlist != BJ_Metadata) { | |
1359 | jbd_lock_bh_state(bh); | |
1360 | J_ASSERT_JH(jh, jh->b_transaction != transaction || | |
1361 | jh->b_jlist == BJ_Metadata); | |
1362 | jbd_unlock_bh_state(bh); | |
1363 | } | |
1364 | goto out; | |
1365 | } | |
1366 | ||
1367 | journal = transaction->t_journal; | |
ad56edad JK |
1368 | jbd_debug(5, "journal_head %p\n", jh); |
1369 | JBUFFER_TRACE(jh, "entry"); | |
470decc6 DK |
1370 | |
1371 | jbd_lock_bh_state(bh); | |
1372 | ||
1373 | if (jh->b_modified == 0) { | |
1374 | /* | |
1375 | * This buffer's got modified and becoming part | |
1376 | * of the transaction. This needs to be done | |
1377 | * once a transaction -bzzz | |
1378 | */ | |
1379 | jh->b_modified = 1; | |
f6c07cad TT |
1380 | if (handle->h_buffer_credits <= 0) { |
1381 | ret = -ENOSPC; | |
1382 | goto out_unlock_bh; | |
1383 | } | |
470decc6 DK |
1384 | handle->h_buffer_credits--; |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * fastpath, to avoid expensive locking. If this buffer is already | |
1389 | * on the running transaction's metadata list there is nothing to do. | |
1390 | * Nobody can take it off again because there is a handle open. | |
1391 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1392 | * result in this test being false, so we go in and take the locks. | |
1393 | */ | |
1394 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1395 | JBUFFER_TRACE(jh, "fastpath"); | |
9ea7a0df TT |
1396 | if (unlikely(jh->b_transaction != |
1397 | journal->j_running_transaction)) { | |
a67c848a | 1398 | printk(KERN_ERR "JBD2: %s: " |
9ea7a0df | 1399 | "jh->b_transaction (%llu, %p, %u) != " |
66a4cb18 | 1400 | "journal->j_running_transaction (%p, %u)\n", |
9ea7a0df TT |
1401 | journal->j_devname, |
1402 | (unsigned long long) bh->b_blocknr, | |
1403 | jh->b_transaction, | |
1404 | jh->b_transaction ? jh->b_transaction->t_tid : 0, | |
1405 | journal->j_running_transaction, | |
1406 | journal->j_running_transaction ? | |
1407 | journal->j_running_transaction->t_tid : 0); | |
1408 | ret = -EINVAL; | |
1409 | } | |
470decc6 DK |
1410 | goto out_unlock_bh; |
1411 | } | |
1412 | ||
1413 | set_buffer_jbddirty(bh); | |
1414 | ||
1415 | /* | |
1416 | * Metadata already on the current transaction list doesn't | |
1417 | * need to be filed. Metadata on another transaction's list must | |
1418 | * be committing, and will be refiled once the commit completes: | |
1419 | * leave it alone for now. | |
1420 | */ | |
1421 | if (jh->b_transaction != transaction) { | |
1422 | JBUFFER_TRACE(jh, "already on other transaction"); | |
66a4cb18 TT |
1423 | if (unlikely(((jh->b_transaction != |
1424 | journal->j_committing_transaction)) || | |
1425 | (jh->b_next_transaction != transaction))) { | |
1426 | printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: " | |
1427 | "bad jh for block %llu: " | |
1428 | "transaction (%p, %u), " | |
1429 | "jh->b_transaction (%p, %u), " | |
1430 | "jh->b_next_transaction (%p, %u), jlist %u\n", | |
9ea7a0df TT |
1431 | journal->j_devname, |
1432 | (unsigned long long) bh->b_blocknr, | |
66a4cb18 | 1433 | transaction, transaction->t_tid, |
9ea7a0df | 1434 | jh->b_transaction, |
66a4cb18 TT |
1435 | jh->b_transaction ? |
1436 | jh->b_transaction->t_tid : 0, | |
9ea7a0df TT |
1437 | jh->b_next_transaction, |
1438 | jh->b_next_transaction ? | |
1439 | jh->b_next_transaction->t_tid : 0, | |
66a4cb18 TT |
1440 | jh->b_jlist); |
1441 | WARN_ON(1); | |
9ea7a0df TT |
1442 | ret = -EINVAL; |
1443 | } | |
470decc6 DK |
1444 | /* And this case is illegal: we can't reuse another |
1445 | * transaction's data buffer, ever. */ | |
1446 | goto out_unlock_bh; | |
1447 | } | |
1448 | ||
1449 | /* That test should have eliminated the following case: */ | |
4019191b | 1450 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1451 | |
1452 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1453 | spin_lock(&journal->j_list_lock); | |
41a5b913 | 1454 | __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata); |
470decc6 DK |
1455 | spin_unlock(&journal->j_list_lock); |
1456 | out_unlock_bh: | |
1457 | jbd_unlock_bh_state(bh); | |
1458 | out: | |
1459 | JBUFFER_TRACE(jh, "exit"); | |
9ea7a0df | 1460 | return ret; |
470decc6 DK |
1461 | } |
1462 | ||
470decc6 | 1463 | /** |
f7f4bccb | 1464 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1465 | * @handle: transaction handle |
1466 | * @bh: bh to 'forget' | |
1467 | * | |
1468 | * We can only do the bforget if there are no commits pending against the | |
1469 | * buffer. If the buffer is dirty in the current running transaction we | |
1470 | * can safely unlink it. | |
1471 | * | |
1472 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1473 | * buffer which came off the hashtable. Check for this. | |
1474 | * | |
1475 | * Decrements bh->b_count by one. | |
1476 | * | |
1477 | * Allow this call even if the handle has aborted --- it may be part of | |
1478 | * the caller's cleanup after an abort. | |
1479 | */ | |
f7f4bccb | 1480 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1481 | { |
1482 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1483 | journal_t *journal; |
470decc6 DK |
1484 | struct journal_head *jh; |
1485 | int drop_reserve = 0; | |
1486 | int err = 0; | |
1dfc3220 | 1487 | int was_modified = 0; |
470decc6 | 1488 | |
41a5b913 TT |
1489 | if (is_handle_aborted(handle)) |
1490 | return -EROFS; | |
1491 | journal = transaction->t_journal; | |
1492 | ||
470decc6 DK |
1493 | BUFFER_TRACE(bh, "entry"); |
1494 | ||
1495 | jbd_lock_bh_state(bh); | |
470decc6 DK |
1496 | |
1497 | if (!buffer_jbd(bh)) | |
1498 | goto not_jbd; | |
1499 | jh = bh2jh(bh); | |
1500 | ||
1501 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1502 | * Don't do any jbd operations, and return an error. */ | |
1503 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1504 | "inconsistent data on disk")) { | |
1505 | err = -EIO; | |
1506 | goto not_jbd; | |
1507 | } | |
1508 | ||
48fc7f7e | 1509 | /* keep track of whether or not this transaction modified us */ |
1dfc3220 JB |
1510 | was_modified = jh->b_modified; |
1511 | ||
470decc6 DK |
1512 | /* |
1513 | * The buffer's going from the transaction, we must drop | |
1514 | * all references -bzzz | |
1515 | */ | |
1516 | jh->b_modified = 0; | |
1517 | ||
41a5b913 | 1518 | if (jh->b_transaction == transaction) { |
470decc6 DK |
1519 | J_ASSERT_JH(jh, !jh->b_frozen_data); |
1520 | ||
1521 | /* If we are forgetting a buffer which is already part | |
1522 | * of this transaction, then we can just drop it from | |
1523 | * the transaction immediately. */ | |
1524 | clear_buffer_dirty(bh); | |
1525 | clear_buffer_jbddirty(bh); | |
1526 | ||
1527 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1528 | ||
1dfc3220 JB |
1529 | /* |
1530 | * we only want to drop a reference if this transaction | |
1531 | * modified the buffer | |
1532 | */ | |
1533 | if (was_modified) | |
1534 | drop_reserve = 1; | |
470decc6 DK |
1535 | |
1536 | /* | |
1537 | * We are no longer going to journal this buffer. | |
1538 | * However, the commit of this transaction is still | |
1539 | * important to the buffer: the delete that we are now | |
1540 | * processing might obsolete an old log entry, so by | |
1541 | * committing, we can satisfy the buffer's checkpoint. | |
1542 | * | |
1543 | * So, if we have a checkpoint on the buffer, we should | |
1544 | * now refile the buffer on our BJ_Forget list so that | |
1545 | * we know to remove the checkpoint after we commit. | |
1546 | */ | |
1547 | ||
0bfea811 | 1548 | spin_lock(&journal->j_list_lock); |
470decc6 | 1549 | if (jh->b_cp_transaction) { |
f7f4bccb MC |
1550 | __jbd2_journal_temp_unlink_buffer(jh); |
1551 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1552 | } else { |
f7f4bccb | 1553 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1554 | if (!buffer_jbd(bh)) { |
1555 | spin_unlock(&journal->j_list_lock); | |
1556 | jbd_unlock_bh_state(bh); | |
1557 | __bforget(bh); | |
1558 | goto drop; | |
1559 | } | |
1560 | } | |
0bfea811 | 1561 | spin_unlock(&journal->j_list_lock); |
470decc6 DK |
1562 | } else if (jh->b_transaction) { |
1563 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1564 | journal->j_committing_transaction)); | |
1565 | /* However, if the buffer is still owned by a prior | |
1566 | * (committing) transaction, we can't drop it yet... */ | |
1567 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
1568 | /* ... but we CAN drop it from the new transaction if we | |
1569 | * have also modified it since the original commit. */ | |
1570 | ||
1571 | if (jh->b_next_transaction) { | |
1572 | J_ASSERT(jh->b_next_transaction == transaction); | |
0bfea811 | 1573 | spin_lock(&journal->j_list_lock); |
470decc6 | 1574 | jh->b_next_transaction = NULL; |
0bfea811 | 1575 | spin_unlock(&journal->j_list_lock); |
1dfc3220 JB |
1576 | |
1577 | /* | |
1578 | * only drop a reference if this transaction modified | |
1579 | * the buffer | |
1580 | */ | |
1581 | if (was_modified) | |
1582 | drop_reserve = 1; | |
470decc6 DK |
1583 | } |
1584 | } | |
1585 | ||
1586 | not_jbd: | |
470decc6 DK |
1587 | jbd_unlock_bh_state(bh); |
1588 | __brelse(bh); | |
1589 | drop: | |
1590 | if (drop_reserve) { | |
1591 | /* no need to reserve log space for this block -bzzz */ | |
1592 | handle->h_buffer_credits++; | |
1593 | } | |
1594 | return err; | |
1595 | } | |
1596 | ||
1597 | /** | |
f7f4bccb | 1598 | * int jbd2_journal_stop() - complete a transaction |
470decc6 DK |
1599 | * @handle: tranaction to complete. |
1600 | * | |
1601 | * All done for a particular handle. | |
1602 | * | |
1603 | * There is not much action needed here. We just return any remaining | |
1604 | * buffer credits to the transaction and remove the handle. The only | |
1605 | * complication is that we need to start a commit operation if the | |
1606 | * filesystem is marked for synchronous update. | |
1607 | * | |
f7f4bccb | 1608 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1609 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1610 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1611 | * transaction began. |
1612 | */ | |
f7f4bccb | 1613 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1614 | { |
1615 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 TT |
1616 | journal_t *journal; |
1617 | int err = 0, wait_for_commit = 0; | |
a51dca9c | 1618 | tid_t tid; |
470decc6 DK |
1619 | pid_t pid; |
1620 | ||
9d506594 LC |
1621 | if (!transaction) { |
1622 | /* | |
1623 | * Handle is already detached from the transaction so | |
1624 | * there is nothing to do other than decrease a refcount, | |
1625 | * or free the handle if refcount drops to zero | |
1626 | */ | |
1627 | if (--handle->h_ref > 0) { | |
1628 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1629 | handle->h_ref); | |
1630 | return err; | |
1631 | } else { | |
1632 | if (handle->h_rsv_handle) | |
1633 | jbd2_free_handle(handle->h_rsv_handle); | |
1634 | goto free_and_exit; | |
1635 | } | |
1636 | } | |
41a5b913 TT |
1637 | journal = transaction->t_journal; |
1638 | ||
470decc6 DK |
1639 | J_ASSERT(journal_current_handle() == handle); |
1640 | ||
1641 | if (is_handle_aborted(handle)) | |
1642 | err = -EIO; | |
41a5b913 | 1643 | else |
a51dca9c | 1644 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
1645 | |
1646 | if (--handle->h_ref > 0) { | |
1647 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1648 | handle->h_ref); | |
1649 | return err; | |
1650 | } | |
1651 | ||
1652 | jbd_debug(4, "Handle %p going down\n", handle); | |
343d9c28 | 1653 | trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev, |
41a5b913 | 1654 | transaction->t_tid, |
343d9c28 TT |
1655 | handle->h_type, handle->h_line_no, |
1656 | jiffies - handle->h_start_jiffies, | |
1657 | handle->h_sync, handle->h_requested_credits, | |
1658 | (handle->h_requested_credits - | |
1659 | handle->h_buffer_credits)); | |
470decc6 DK |
1660 | |
1661 | /* | |
1662 | * Implement synchronous transaction batching. If the handle | |
1663 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1664 | * yield and let another thread piggyback onto this |
1665 | * transaction. Keep doing that while new threads continue to | |
1666 | * arrive. It doesn't cost much - we're about to run a commit | |
1667 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1668 | * operations by 30x or more... | |
1669 | * | |
1670 | * We try and optimize the sleep time against what the | |
1671 | * underlying disk can do, instead of having a static sleep | |
1672 | * time. This is useful for the case where our storage is so | |
1673 | * fast that it is more optimal to go ahead and force a flush | |
1674 | * and wait for the transaction to be committed than it is to | |
1675 | * wait for an arbitrary amount of time for new writers to | |
1676 | * join the transaction. We achieve this by measuring how | |
1677 | * long it takes to commit a transaction, and compare it with | |
1678 | * how long this transaction has been running, and if run time | |
1679 | * < commit time then we sleep for the delta and commit. This | |
1680 | * greatly helps super fast disks that would see slowdowns as | |
1681 | * more threads started doing fsyncs. | |
470decc6 | 1682 | * |
e07f7183 JB |
1683 | * But don't do this if this process was the most recent one |
1684 | * to perform a synchronous write. We do this to detect the | |
1685 | * case where a single process is doing a stream of sync | |
1686 | * writes. No point in waiting for joiners in that case. | |
5dd21424 ES |
1687 | * |
1688 | * Setting max_batch_time to 0 disables this completely. | |
470decc6 DK |
1689 | */ |
1690 | pid = current->pid; | |
5dd21424 ES |
1691 | if (handle->h_sync && journal->j_last_sync_writer != pid && |
1692 | journal->j_max_batch_time) { | |
e07f7183 JB |
1693 | u64 commit_time, trans_time; |
1694 | ||
470decc6 | 1695 | journal->j_last_sync_writer = pid; |
e07f7183 | 1696 | |
a931da6a | 1697 | read_lock(&journal->j_state_lock); |
e07f7183 | 1698 | commit_time = journal->j_average_commit_time; |
a931da6a | 1699 | read_unlock(&journal->j_state_lock); |
e07f7183 JB |
1700 | |
1701 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1702 | transaction->t_start_time)); | |
1703 | ||
30773840 TT |
1704 | commit_time = max_t(u64, commit_time, |
1705 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1706 | commit_time = min_t(u64, commit_time, |
30773840 | 1707 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1708 | |
1709 | if (trans_time < commit_time) { | |
1710 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1711 | commit_time); | |
1712 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1713 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1714 | } | |
470decc6 DK |
1715 | } |
1716 | ||
7058548c TT |
1717 | if (handle->h_sync) |
1718 | transaction->t_synchronous_commit = 1; | |
470decc6 | 1719 | current->journal_info = NULL; |
a51dca9c TT |
1720 | atomic_sub(handle->h_buffer_credits, |
1721 | &transaction->t_outstanding_credits); | |
470decc6 DK |
1722 | |
1723 | /* | |
1724 | * If the handle is marked SYNC, we need to set another commit | |
1725 | * going! We also want to force a commit if the current | |
1726 | * transaction is occupying too much of the log, or if the | |
1727 | * transaction is too old now. | |
1728 | */ | |
1729 | if (handle->h_sync || | |
a51dca9c TT |
1730 | (atomic_read(&transaction->t_outstanding_credits) > |
1731 | journal->j_max_transaction_buffers) || | |
1732 | time_after_eq(jiffies, transaction->t_expires)) { | |
470decc6 DK |
1733 | /* Do this even for aborted journals: an abort still |
1734 | * completes the commit thread, it just doesn't write | |
1735 | * anything to disk. */ | |
470decc6 | 1736 | |
470decc6 DK |
1737 | jbd_debug(2, "transaction too old, requesting commit for " |
1738 | "handle %p\n", handle); | |
1739 | /* This is non-blocking */ | |
c35a56a0 | 1740 | jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1741 | |
1742 | /* | |
f7f4bccb | 1743 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1744 | * to wait for the commit to complete. |
1745 | */ | |
1746 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
a51dca9c | 1747 | wait_for_commit = 1; |
470decc6 DK |
1748 | } |
1749 | ||
a51dca9c TT |
1750 | /* |
1751 | * Once we drop t_updates, if it goes to zero the transaction | |
25985edc | 1752 | * could start committing on us and eventually disappear. So |
a51dca9c TT |
1753 | * once we do this, we must not dereference transaction |
1754 | * pointer again. | |
1755 | */ | |
1756 | tid = transaction->t_tid; | |
1757 | if (atomic_dec_and_test(&transaction->t_updates)) { | |
1758 | wake_up(&journal->j_wait_updates); | |
1759 | if (journal->j_barrier_count) | |
1760 | wake_up(&journal->j_wait_transaction_locked); | |
1761 | } | |
1762 | ||
1763 | if (wait_for_commit) | |
1764 | err = jbd2_log_wait_commit(journal, tid); | |
1765 | ||
3295f0ef | 1766 | lock_map_release(&handle->h_lockdep_map); |
7b751066 | 1767 | |
8f7d89f3 JK |
1768 | if (handle->h_rsv_handle) |
1769 | jbd2_journal_free_reserved(handle->h_rsv_handle); | |
41a5b913 | 1770 | free_and_exit: |
af1e76d6 | 1771 | jbd2_free_handle(handle); |
470decc6 DK |
1772 | return err; |
1773 | } | |
1774 | ||
470decc6 DK |
1775 | /* |
1776 | * | |
1777 | * List management code snippets: various functions for manipulating the | |
1778 | * transaction buffer lists. | |
1779 | * | |
1780 | */ | |
1781 | ||
1782 | /* | |
1783 | * Append a buffer to a transaction list, given the transaction's list head | |
1784 | * pointer. | |
1785 | * | |
1786 | * j_list_lock is held. | |
1787 | * | |
1788 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1789 | */ | |
1790 | ||
1791 | static inline void | |
1792 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1793 | { | |
1794 | if (!*list) { | |
1795 | jh->b_tnext = jh->b_tprev = jh; | |
1796 | *list = jh; | |
1797 | } else { | |
1798 | /* Insert at the tail of the list to preserve order */ | |
1799 | struct journal_head *first = *list, *last = first->b_tprev; | |
1800 | jh->b_tprev = last; | |
1801 | jh->b_tnext = first; | |
1802 | last->b_tnext = first->b_tprev = jh; | |
1803 | } | |
1804 | } | |
1805 | ||
1806 | /* | |
1807 | * Remove a buffer from a transaction list, given the transaction's list | |
1808 | * head pointer. | |
1809 | * | |
1810 | * Called with j_list_lock held, and the journal may not be locked. | |
1811 | * | |
1812 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1813 | */ | |
1814 | ||
1815 | static inline void | |
1816 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
1817 | { | |
1818 | if (*list == jh) { | |
1819 | *list = jh->b_tnext; | |
1820 | if (*list == jh) | |
1821 | *list = NULL; | |
1822 | } | |
1823 | jh->b_tprev->b_tnext = jh->b_tnext; | |
1824 | jh->b_tnext->b_tprev = jh->b_tprev; | |
1825 | } | |
1826 | ||
1827 | /* | |
1828 | * Remove a buffer from the appropriate transaction list. | |
1829 | * | |
1830 | * Note that this function can *change* the value of | |
f5113eff JK |
1831 | * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or |
1832 | * t_reserved_list. If the caller is holding onto a copy of one of these | |
1833 | * pointers, it could go bad. Generally the caller needs to re-read the | |
1834 | * pointer from the transaction_t. | |
470decc6 | 1835 | * |
5bebccf9 | 1836 | * Called under j_list_lock. |
470decc6 | 1837 | */ |
5bebccf9 | 1838 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
1839 | { |
1840 | struct journal_head **list = NULL; | |
1841 | transaction_t *transaction; | |
1842 | struct buffer_head *bh = jh2bh(jh); | |
1843 | ||
1844 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1845 | transaction = jh->b_transaction; | |
1846 | if (transaction) | |
1847 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1848 | ||
1849 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1850 | if (jh->b_jlist != BJ_None) | |
4019191b | 1851 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
1852 | |
1853 | switch (jh->b_jlist) { | |
1854 | case BJ_None: | |
1855 | return; | |
470decc6 DK |
1856 | case BJ_Metadata: |
1857 | transaction->t_nr_buffers--; | |
1858 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
1859 | list = &transaction->t_buffers; | |
1860 | break; | |
1861 | case BJ_Forget: | |
1862 | list = &transaction->t_forget; | |
1863 | break; | |
470decc6 DK |
1864 | case BJ_Shadow: |
1865 | list = &transaction->t_shadow_list; | |
1866 | break; | |
470decc6 DK |
1867 | case BJ_Reserved: |
1868 | list = &transaction->t_reserved_list; | |
1869 | break; | |
470decc6 DK |
1870 | } |
1871 | ||
1872 | __blist_del_buffer(list, jh); | |
1873 | jh->b_jlist = BJ_None; | |
1874 | if (test_clear_buffer_jbddirty(bh)) | |
1875 | mark_buffer_dirty(bh); /* Expose it to the VM */ | |
1876 | } | |
1877 | ||
de1b7941 JK |
1878 | /* |
1879 | * Remove buffer from all transactions. | |
1880 | * | |
1881 | * Called with bh_state lock and j_list_lock | |
1882 | * | |
1883 | * jh and bh may be already freed when this function returns. | |
1884 | */ | |
1885 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh) | |
470decc6 | 1886 | { |
f7f4bccb | 1887 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 | 1888 | jh->b_transaction = NULL; |
de1b7941 | 1889 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1890 | } |
1891 | ||
f7f4bccb | 1892 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 | 1893 | { |
de1b7941 JK |
1894 | struct buffer_head *bh = jh2bh(jh); |
1895 | ||
1896 | /* Get reference so that buffer cannot be freed before we unlock it */ | |
1897 | get_bh(bh); | |
1898 | jbd_lock_bh_state(bh); | |
470decc6 | 1899 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 1900 | __jbd2_journal_unfile_buffer(jh); |
470decc6 | 1901 | spin_unlock(&journal->j_list_lock); |
de1b7941 JK |
1902 | jbd_unlock_bh_state(bh); |
1903 | __brelse(bh); | |
470decc6 DK |
1904 | } |
1905 | ||
1906 | /* | |
f7f4bccb | 1907 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 DK |
1908 | * |
1909 | * Called under jbd_lock_bh_state(bh) | |
1910 | */ | |
1911 | static void | |
1912 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
1913 | { | |
1914 | struct journal_head *jh; | |
1915 | ||
1916 | jh = bh2jh(bh); | |
1917 | ||
1918 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
1919 | goto out; | |
1920 | ||
d2eb0b99 | 1921 | if (jh->b_next_transaction != NULL || jh->b_transaction != NULL) |
470decc6 DK |
1922 | goto out; |
1923 | ||
1924 | spin_lock(&journal->j_list_lock); | |
d2eb0b99 | 1925 | if (jh->b_cp_transaction != NULL) { |
470decc6 | 1926 | /* written-back checkpointed metadata buffer */ |
c254c9ec JK |
1927 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
1928 | __jbd2_journal_remove_checkpoint(jh); | |
470decc6 DK |
1929 | } |
1930 | spin_unlock(&journal->j_list_lock); | |
1931 | out: | |
1932 | return; | |
1933 | } | |
1934 | ||
470decc6 | 1935 | /** |
f7f4bccb | 1936 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 DK |
1937 | * @journal: journal for operation |
1938 | * @page: to try and free | |
530576bb MC |
1939 | * @gfp_mask: we use the mask to detect how hard should we try to release |
1940 | * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to | |
1941 | * release the buffers. | |
470decc6 DK |
1942 | * |
1943 | * | |
1944 | * For all the buffers on this page, | |
1945 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
1946 | * so try_to_free_buffers() can reap them. | |
1947 | * | |
1948 | * This function returns non-zero if we wish try_to_free_buffers() | |
1949 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
1950 | * We also do it if the page has locked or dirty buffers and the caller wants | |
1951 | * us to perform sync or async writeout. | |
1952 | * | |
1953 | * This complicates JBD locking somewhat. We aren't protected by the | |
1954 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 1955 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
1956 | * |
1957 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
1958 | * who looks at t_datalist needs to lock against this function. | |
1959 | * | |
f7f4bccb MC |
1960 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
1961 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
1962 | * will come out of the lock with the buffer dirty, which makes it |
1963 | * ineligible for release here. | |
1964 | * | |
1965 | * Who else is affected by this? hmm... Really the only contender | |
1966 | * is do_get_write_access() - it could be looking at the buffer while | |
1967 | * journal_try_to_free_buffer() is changing its state. But that | |
1968 | * cannot happen because we never reallocate freed data as metadata | |
1969 | * while the data is part of a transaction. Yes? | |
530576bb MC |
1970 | * |
1971 | * Return 0 on failure, 1 on success | |
470decc6 | 1972 | */ |
f7f4bccb | 1973 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
530576bb | 1974 | struct page *page, gfp_t gfp_mask) |
470decc6 DK |
1975 | { |
1976 | struct buffer_head *head; | |
1977 | struct buffer_head *bh; | |
1978 | int ret = 0; | |
1979 | ||
1980 | J_ASSERT(PageLocked(page)); | |
1981 | ||
1982 | head = page_buffers(page); | |
1983 | bh = head; | |
1984 | do { | |
1985 | struct journal_head *jh; | |
1986 | ||
1987 | /* | |
1988 | * We take our own ref against the journal_head here to avoid | |
1989 | * having to add tons of locking around each instance of | |
530576bb | 1990 | * jbd2_journal_put_journal_head(). |
470decc6 | 1991 | */ |
f7f4bccb | 1992 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1993 | if (!jh) |
1994 | continue; | |
1995 | ||
1996 | jbd_lock_bh_state(bh); | |
1997 | __journal_try_to_free_buffer(journal, bh); | |
f7f4bccb | 1998 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1999 | jbd_unlock_bh_state(bh); |
2000 | if (buffer_jbd(bh)) | |
2001 | goto busy; | |
2002 | } while ((bh = bh->b_this_page) != head); | |
530576bb | 2003 | |
470decc6 | 2004 | ret = try_to_free_buffers(page); |
530576bb | 2005 | |
470decc6 DK |
2006 | busy: |
2007 | return ret; | |
2008 | } | |
2009 | ||
2010 | /* | |
2011 | * This buffer is no longer needed. If it is on an older transaction's | |
2012 | * checkpoint list we need to record it on this transaction's forget list | |
2013 | * to pin this buffer (and hence its checkpointing transaction) down until | |
2014 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
2015 | * release it. | |
2016 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
2017 | * | |
2018 | * Called under j_list_lock. | |
2019 | * | |
2020 | * Called under jbd_lock_bh_state(bh). | |
2021 | */ | |
2022 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
2023 | { | |
2024 | int may_free = 1; | |
2025 | struct buffer_head *bh = jh2bh(jh); | |
2026 | ||
470decc6 DK |
2027 | if (jh->b_cp_transaction) { |
2028 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
de1b7941 | 2029 | __jbd2_journal_temp_unlink_buffer(jh); |
f91d1d04 JK |
2030 | /* |
2031 | * We don't want to write the buffer anymore, clear the | |
2032 | * bit so that we don't confuse checks in | |
2033 | * __journal_file_buffer | |
2034 | */ | |
2035 | clear_buffer_dirty(bh); | |
f7f4bccb | 2036 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
2037 | may_free = 0; |
2038 | } else { | |
2039 | JBUFFER_TRACE(jh, "on running transaction"); | |
de1b7941 | 2040 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2041 | } |
2042 | return may_free; | |
2043 | } | |
2044 | ||
2045 | /* | |
f7f4bccb | 2046 | * jbd2_journal_invalidatepage |
470decc6 DK |
2047 | * |
2048 | * This code is tricky. It has a number of cases to deal with. | |
2049 | * | |
2050 | * There are two invariants which this code relies on: | |
2051 | * | |
2052 | * i_size must be updated on disk before we start calling invalidatepage on the | |
2053 | * data. | |
2054 | * | |
2055 | * This is done in ext3 by defining an ext3_setattr method which | |
2056 | * updates i_size before truncate gets going. By maintaining this | |
2057 | * invariant, we can be sure that it is safe to throw away any buffers | |
2058 | * attached to the current transaction: once the transaction commits, | |
2059 | * we know that the data will not be needed. | |
2060 | * | |
2061 | * Note however that we can *not* throw away data belonging to the | |
2062 | * previous, committing transaction! | |
2063 | * | |
2064 | * Any disk blocks which *are* part of the previous, committing | |
2065 | * transaction (and which therefore cannot be discarded immediately) are | |
2066 | * not going to be reused in the new running transaction | |
2067 | * | |
2068 | * The bitmap committed_data images guarantee this: any block which is | |
2069 | * allocated in one transaction and removed in the next will be marked | |
2070 | * as in-use in the committed_data bitmap, so cannot be reused until | |
2071 | * the next transaction to delete the block commits. This means that | |
2072 | * leaving committing buffers dirty is quite safe: the disk blocks | |
2073 | * cannot be reallocated to a different file and so buffer aliasing is | |
2074 | * not possible. | |
2075 | * | |
2076 | * | |
2077 | * The above applies mainly to ordered data mode. In writeback mode we | |
2078 | * don't make guarantees about the order in which data hits disk --- in | |
2079 | * particular we don't guarantee that new dirty data is flushed before | |
2080 | * transaction commit --- so it is always safe just to discard data | |
2081 | * immediately in that mode. --sct | |
2082 | */ | |
2083 | ||
2084 | /* | |
2085 | * The journal_unmap_buffer helper function returns zero if the buffer | |
2086 | * concerned remains pinned as an anonymous buffer belonging to an older | |
2087 | * transaction. | |
2088 | * | |
2089 | * We're outside-transaction here. Either or both of j_running_transaction | |
2090 | * and j_committing_transaction may be NULL. | |
2091 | */ | |
b794e7a6 JK |
2092 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh, |
2093 | int partial_page) | |
470decc6 DK |
2094 | { |
2095 | transaction_t *transaction; | |
2096 | struct journal_head *jh; | |
2097 | int may_free = 1; | |
470decc6 DK |
2098 | |
2099 | BUFFER_TRACE(bh, "entry"); | |
2100 | ||
2101 | /* | |
2102 | * It is safe to proceed here without the j_list_lock because the | |
2103 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
2104 | * holding the page lock. --sct | |
2105 | */ | |
2106 | ||
2107 | if (!buffer_jbd(bh)) | |
2108 | goto zap_buffer_unlocked; | |
2109 | ||
87c89c23 | 2110 | /* OK, we have data buffer in journaled mode */ |
a931da6a | 2111 | write_lock(&journal->j_state_lock); |
470decc6 DK |
2112 | jbd_lock_bh_state(bh); |
2113 | spin_lock(&journal->j_list_lock); | |
2114 | ||
f7f4bccb | 2115 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
2116 | if (!jh) |
2117 | goto zap_buffer_no_jh; | |
2118 | ||
ba869023 | 2119 | /* |
2120 | * We cannot remove the buffer from checkpoint lists until the | |
2121 | * transaction adding inode to orphan list (let's call it T) | |
2122 | * is committed. Otherwise if the transaction changing the | |
2123 | * buffer would be cleaned from the journal before T is | |
2124 | * committed, a crash will cause that the correct contents of | |
2125 | * the buffer will be lost. On the other hand we have to | |
2126 | * clear the buffer dirty bit at latest at the moment when the | |
2127 | * transaction marking the buffer as freed in the filesystem | |
2128 | * structures is committed because from that moment on the | |
b794e7a6 | 2129 | * block can be reallocated and used by a different page. |
ba869023 | 2130 | * Since the block hasn't been freed yet but the inode has |
2131 | * already been added to orphan list, it is safe for us to add | |
2132 | * the buffer to BJ_Forget list of the newest transaction. | |
b794e7a6 JK |
2133 | * |
2134 | * Also we have to clear buffer_mapped flag of a truncated buffer | |
2135 | * because the buffer_head may be attached to the page straddling | |
2136 | * i_size (can happen only when blocksize < pagesize) and thus the | |
2137 | * buffer_head can be reused when the file is extended again. So we end | |
2138 | * up keeping around invalidated buffers attached to transactions' | |
2139 | * BJ_Forget list just to stop checkpointing code from cleaning up | |
2140 | * the transaction this buffer was modified in. | |
ba869023 | 2141 | */ |
470decc6 DK |
2142 | transaction = jh->b_transaction; |
2143 | if (transaction == NULL) { | |
2144 | /* First case: not on any transaction. If it | |
2145 | * has no checkpoint link, then we can zap it: | |
2146 | * it's a writeback-mode buffer so we don't care | |
2147 | * if it hits disk safely. */ | |
2148 | if (!jh->b_cp_transaction) { | |
2149 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
2150 | goto zap_buffer; | |
2151 | } | |
2152 | ||
2153 | if (!buffer_dirty(bh)) { | |
2154 | /* bdflush has written it. We can drop it now */ | |
2155 | goto zap_buffer; | |
2156 | } | |
2157 | ||
2158 | /* OK, it must be in the journal but still not | |
2159 | * written fully to disk: it's metadata or | |
2160 | * journaled data... */ | |
2161 | ||
2162 | if (journal->j_running_transaction) { | |
2163 | /* ... and once the current transaction has | |
2164 | * committed, the buffer won't be needed any | |
2165 | * longer. */ | |
2166 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
b794e7a6 | 2167 | may_free = __dispose_buffer(jh, |
470decc6 | 2168 | journal->j_running_transaction); |
b794e7a6 | 2169 | goto zap_buffer; |
470decc6 DK |
2170 | } else { |
2171 | /* There is no currently-running transaction. So the | |
2172 | * orphan record which we wrote for this file must have | |
2173 | * passed into commit. We must attach this buffer to | |
2174 | * the committing transaction, if it exists. */ | |
2175 | if (journal->j_committing_transaction) { | |
2176 | JBUFFER_TRACE(jh, "give to committing trans"); | |
b794e7a6 | 2177 | may_free = __dispose_buffer(jh, |
470decc6 | 2178 | journal->j_committing_transaction); |
b794e7a6 | 2179 | goto zap_buffer; |
470decc6 DK |
2180 | } else { |
2181 | /* The orphan record's transaction has | |
2182 | * committed. We can cleanse this buffer */ | |
2183 | clear_buffer_jbddirty(bh); | |
2184 | goto zap_buffer; | |
2185 | } | |
2186 | } | |
2187 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 2188 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 2189 | /* |
ba869023 | 2190 | * The buffer is committing, we simply cannot touch |
b794e7a6 JK |
2191 | * it. If the page is straddling i_size we have to wait |
2192 | * for commit and try again. | |
2193 | */ | |
2194 | if (partial_page) { | |
b794e7a6 JK |
2195 | jbd2_journal_put_journal_head(jh); |
2196 | spin_unlock(&journal->j_list_lock); | |
2197 | jbd_unlock_bh_state(bh); | |
2198 | write_unlock(&journal->j_state_lock); | |
53e87268 | 2199 | return -EBUSY; |
b794e7a6 JK |
2200 | } |
2201 | /* | |
2202 | * OK, buffer won't be reachable after truncate. We just set | |
2203 | * j_next_transaction to the running transaction (if there is | |
2204 | * one) and mark buffer as freed so that commit code knows it | |
2205 | * should clear dirty bits when it is done with the buffer. | |
ba869023 | 2206 | */ |
470decc6 | 2207 | set_buffer_freed(bh); |
ba869023 | 2208 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
2209 | jh->b_next_transaction = journal->j_running_transaction; | |
f7f4bccb | 2210 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2211 | spin_unlock(&journal->j_list_lock); |
2212 | jbd_unlock_bh_state(bh); | |
a931da6a | 2213 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2214 | return 0; |
2215 | } else { | |
2216 | /* Good, the buffer belongs to the running transaction. | |
2217 | * We are writing our own transaction's data, not any | |
2218 | * previous one's, so it is safe to throw it away | |
2219 | * (remember that we expect the filesystem to have set | |
2220 | * i_size already for this truncate so recovery will not | |
2221 | * expose the disk blocks we are discarding here.) */ | |
2222 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 2223 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
2224 | may_free = __dispose_buffer(jh, transaction); |
2225 | } | |
2226 | ||
2227 | zap_buffer: | |
b794e7a6 JK |
2228 | /* |
2229 | * This is tricky. Although the buffer is truncated, it may be reused | |
2230 | * if blocksize < pagesize and it is attached to the page straddling | |
2231 | * EOF. Since the buffer might have been added to BJ_Forget list of the | |
2232 | * running transaction, journal_get_write_access() won't clear | |
2233 | * b_modified and credit accounting gets confused. So clear b_modified | |
2234 | * here. | |
2235 | */ | |
2236 | jh->b_modified = 0; | |
f7f4bccb | 2237 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2238 | zap_buffer_no_jh: |
2239 | spin_unlock(&journal->j_list_lock); | |
2240 | jbd_unlock_bh_state(bh); | |
a931da6a | 2241 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2242 | zap_buffer_unlocked: |
2243 | clear_buffer_dirty(bh); | |
2244 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
2245 | clear_buffer_mapped(bh); | |
2246 | clear_buffer_req(bh); | |
2247 | clear_buffer_new(bh); | |
15291164 ES |
2248 | clear_buffer_delay(bh); |
2249 | clear_buffer_unwritten(bh); | |
470decc6 DK |
2250 | bh->b_bdev = NULL; |
2251 | return may_free; | |
2252 | } | |
2253 | ||
2254 | /** | |
f7f4bccb | 2255 | * void jbd2_journal_invalidatepage() |
470decc6 DK |
2256 | * @journal: journal to use for flush... |
2257 | * @page: page to flush | |
259709b0 LC |
2258 | * @offset: start of the range to invalidate |
2259 | * @length: length of the range to invalidate | |
470decc6 | 2260 | * |
259709b0 LC |
2261 | * Reap page buffers containing data after in the specified range in page. |
2262 | * Can return -EBUSY if buffers are part of the committing transaction and | |
2263 | * the page is straddling i_size. Caller then has to wait for current commit | |
2264 | * and try again. | |
470decc6 | 2265 | */ |
53e87268 JK |
2266 | int jbd2_journal_invalidatepage(journal_t *journal, |
2267 | struct page *page, | |
259709b0 LC |
2268 | unsigned int offset, |
2269 | unsigned int length) | |
470decc6 DK |
2270 | { |
2271 | struct buffer_head *head, *bh, *next; | |
259709b0 | 2272 | unsigned int stop = offset + length; |
470decc6 | 2273 | unsigned int curr_off = 0; |
259709b0 | 2274 | int partial_page = (offset || length < PAGE_CACHE_SIZE); |
470decc6 | 2275 | int may_free = 1; |
53e87268 | 2276 | int ret = 0; |
470decc6 DK |
2277 | |
2278 | if (!PageLocked(page)) | |
2279 | BUG(); | |
2280 | if (!page_has_buffers(page)) | |
53e87268 | 2281 | return 0; |
470decc6 | 2282 | |
259709b0 LC |
2283 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); |
2284 | ||
470decc6 DK |
2285 | /* We will potentially be playing with lists other than just the |
2286 | * data lists (especially for journaled data mode), so be | |
2287 | * cautious in our locking. */ | |
2288 | ||
2289 | head = bh = page_buffers(page); | |
2290 | do { | |
2291 | unsigned int next_off = curr_off + bh->b_size; | |
2292 | next = bh->b_this_page; | |
2293 | ||
259709b0 LC |
2294 | if (next_off > stop) |
2295 | return 0; | |
2296 | ||
470decc6 DK |
2297 | if (offset <= curr_off) { |
2298 | /* This block is wholly outside the truncation point */ | |
2299 | lock_buffer(bh); | |
259709b0 | 2300 | ret = journal_unmap_buffer(journal, bh, partial_page); |
470decc6 | 2301 | unlock_buffer(bh); |
53e87268 JK |
2302 | if (ret < 0) |
2303 | return ret; | |
2304 | may_free &= ret; | |
470decc6 DK |
2305 | } |
2306 | curr_off = next_off; | |
2307 | bh = next; | |
2308 | ||
2309 | } while (bh != head); | |
2310 | ||
259709b0 | 2311 | if (!partial_page) { |
470decc6 DK |
2312 | if (may_free && try_to_free_buffers(page)) |
2313 | J_ASSERT(!page_has_buffers(page)); | |
2314 | } | |
53e87268 | 2315 | return 0; |
470decc6 DK |
2316 | } |
2317 | ||
2318 | /* | |
2319 | * File a buffer on the given transaction list. | |
2320 | */ | |
f7f4bccb | 2321 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2322 | transaction_t *transaction, int jlist) |
2323 | { | |
2324 | struct journal_head **list = NULL; | |
2325 | int was_dirty = 0; | |
2326 | struct buffer_head *bh = jh2bh(jh); | |
2327 | ||
2328 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2329 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
2330 | ||
2331 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
2332 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 2333 | jh->b_transaction == NULL); |
470decc6 DK |
2334 | |
2335 | if (jh->b_transaction && jh->b_jlist == jlist) | |
2336 | return; | |
2337 | ||
470decc6 DK |
2338 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
2339 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
2340 | /* |
2341 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
2342 | * instead of buffer_dirty. We should not see a dirty bit set | |
2343 | * here because we clear it in do_get_write_access but e.g. | |
2344 | * tune2fs can modify the sb and set the dirty bit at any time | |
2345 | * so we try to gracefully handle that. | |
2346 | */ | |
2347 | if (buffer_dirty(bh)) | |
2348 | warn_dirty_buffer(bh); | |
470decc6 DK |
2349 | if (test_clear_buffer_dirty(bh) || |
2350 | test_clear_buffer_jbddirty(bh)) | |
2351 | was_dirty = 1; | |
2352 | } | |
2353 | ||
2354 | if (jh->b_transaction) | |
f7f4bccb | 2355 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2356 | else |
2357 | jbd2_journal_grab_journal_head(bh); | |
470decc6 DK |
2358 | jh->b_transaction = transaction; |
2359 | ||
2360 | switch (jlist) { | |
2361 | case BJ_None: | |
2362 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
2363 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
2364 | return; | |
470decc6 DK |
2365 | case BJ_Metadata: |
2366 | transaction->t_nr_buffers++; | |
2367 | list = &transaction->t_buffers; | |
2368 | break; | |
2369 | case BJ_Forget: | |
2370 | list = &transaction->t_forget; | |
2371 | break; | |
470decc6 DK |
2372 | case BJ_Shadow: |
2373 | list = &transaction->t_shadow_list; | |
2374 | break; | |
470decc6 DK |
2375 | case BJ_Reserved: |
2376 | list = &transaction->t_reserved_list; | |
2377 | break; | |
470decc6 DK |
2378 | } |
2379 | ||
2380 | __blist_add_buffer(list, jh); | |
2381 | jh->b_jlist = jlist; | |
2382 | ||
2383 | if (was_dirty) | |
2384 | set_buffer_jbddirty(bh); | |
2385 | } | |
2386 | ||
f7f4bccb | 2387 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2388 | transaction_t *transaction, int jlist) |
2389 | { | |
2390 | jbd_lock_bh_state(jh2bh(jh)); | |
2391 | spin_lock(&transaction->t_journal->j_list_lock); | |
f7f4bccb | 2392 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 DK |
2393 | spin_unlock(&transaction->t_journal->j_list_lock); |
2394 | jbd_unlock_bh_state(jh2bh(jh)); | |
2395 | } | |
2396 | ||
2397 | /* | |
2398 | * Remove a buffer from its current buffer list in preparation for | |
2399 | * dropping it from its current transaction entirely. If the buffer has | |
2400 | * already started to be used by a subsequent transaction, refile the | |
2401 | * buffer on that transaction's metadata list. | |
2402 | * | |
de1b7941 | 2403 | * Called under j_list_lock |
470decc6 | 2404 | * Called under jbd_lock_bh_state(jh2bh(jh)) |
de1b7941 JK |
2405 | * |
2406 | * jh and bh may be already free when this function returns | |
470decc6 | 2407 | */ |
f7f4bccb | 2408 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2409 | { |
ba869023 | 2410 | int was_dirty, jlist; |
470decc6 DK |
2411 | struct buffer_head *bh = jh2bh(jh); |
2412 | ||
2413 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2414 | if (jh->b_transaction) | |
2415 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2416 | ||
2417 | /* If the buffer is now unused, just drop it. */ | |
2418 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2419 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2420 | return; |
2421 | } | |
2422 | ||
2423 | /* | |
2424 | * It has been modified by a later transaction: add it to the new | |
2425 | * transaction's metadata list. | |
2426 | */ | |
2427 | ||
2428 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2429 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2430 | /* |
2431 | * We set b_transaction here because b_next_transaction will inherit | |
2432 | * our jh reference and thus __jbd2_journal_file_buffer() must not | |
2433 | * take a new one. | |
2434 | */ | |
470decc6 DK |
2435 | jh->b_transaction = jh->b_next_transaction; |
2436 | jh->b_next_transaction = NULL; | |
ba869023 | 2437 | if (buffer_freed(bh)) |
2438 | jlist = BJ_Forget; | |
2439 | else if (jh->b_modified) | |
2440 | jlist = BJ_Metadata; | |
2441 | else | |
2442 | jlist = BJ_Reserved; | |
2443 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2444 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2445 | ||
2446 | if (was_dirty) | |
2447 | set_buffer_jbddirty(bh); | |
2448 | } | |
2449 | ||
2450 | /* | |
de1b7941 JK |
2451 | * __jbd2_journal_refile_buffer() with necessary locking added. We take our |
2452 | * bh reference so that we can safely unlock bh. | |
2453 | * | |
2454 | * The jh and bh may be freed by this call. | |
470decc6 | 2455 | */ |
f7f4bccb | 2456 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
2457 | { |
2458 | struct buffer_head *bh = jh2bh(jh); | |
2459 | ||
de1b7941 JK |
2460 | /* Get reference so that buffer cannot be freed before we unlock it */ |
2461 | get_bh(bh); | |
470decc6 DK |
2462 | jbd_lock_bh_state(bh); |
2463 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 2464 | __jbd2_journal_refile_buffer(jh); |
470decc6 | 2465 | jbd_unlock_bh_state(bh); |
470decc6 DK |
2466 | spin_unlock(&journal->j_list_lock); |
2467 | __brelse(bh); | |
2468 | } | |
c851ed54 JK |
2469 | |
2470 | /* | |
2471 | * File inode in the inode list of the handle's transaction | |
2472 | */ | |
2473 | int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode) | |
2474 | { | |
2475 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 2476 | journal_t *journal; |
c851ed54 JK |
2477 | |
2478 | if (is_handle_aborted(handle)) | |
41a5b913 TT |
2479 | return -EROFS; |
2480 | journal = transaction->t_journal; | |
c851ed54 JK |
2481 | |
2482 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, | |
2483 | transaction->t_tid); | |
2484 | ||
2485 | /* | |
2486 | * First check whether inode isn't already on the transaction's | |
2487 | * lists without taking the lock. Note that this check is safe | |
2488 | * without the lock as we cannot race with somebody removing inode | |
2489 | * from the transaction. The reason is that we remove inode from the | |
2490 | * transaction only in journal_release_jbd_inode() and when we commit | |
2491 | * the transaction. We are guarded from the first case by holding | |
2492 | * a reference to the inode. We are safe against the second case | |
2493 | * because if jinode->i_transaction == transaction, commit code | |
2494 | * cannot touch the transaction because we hold reference to it, | |
2495 | * and if jinode->i_next_transaction == transaction, commit code | |
2496 | * will only file the inode where we want it. | |
2497 | */ | |
2498 | if (jinode->i_transaction == transaction || | |
2499 | jinode->i_next_transaction == transaction) | |
2500 | return 0; | |
2501 | ||
2502 | spin_lock(&journal->j_list_lock); | |
2503 | ||
2504 | if (jinode->i_transaction == transaction || | |
2505 | jinode->i_next_transaction == transaction) | |
2506 | goto done; | |
2507 | ||
81be12c8 JK |
2508 | /* |
2509 | * We only ever set this variable to 1 so the test is safe. Since | |
2510 | * t_need_data_flush is likely to be set, we do the test to save some | |
2511 | * cacheline bouncing | |
2512 | */ | |
2513 | if (!transaction->t_need_data_flush) | |
2514 | transaction->t_need_data_flush = 1; | |
c851ed54 JK |
2515 | /* On some different transaction's list - should be |
2516 | * the committing one */ | |
2517 | if (jinode->i_transaction) { | |
2518 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2519 | J_ASSERT(jinode->i_transaction == | |
2520 | journal->j_committing_transaction); | |
2521 | jinode->i_next_transaction = transaction; | |
2522 | goto done; | |
2523 | } | |
2524 | /* Not on any transaction list... */ | |
2525 | J_ASSERT(!jinode->i_next_transaction); | |
2526 | jinode->i_transaction = transaction; | |
2527 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2528 | done: | |
2529 | spin_unlock(&journal->j_list_lock); | |
2530 | ||
2531 | return 0; | |
2532 | } | |
2533 | ||
2534 | /* | |
7f5aa215 JK |
2535 | * File truncate and transaction commit interact with each other in a |
2536 | * non-trivial way. If a transaction writing data block A is | |
2537 | * committing, we cannot discard the data by truncate until we have | |
2538 | * written them. Otherwise if we crashed after the transaction with | |
2539 | * write has committed but before the transaction with truncate has | |
2540 | * committed, we could see stale data in block A. This function is a | |
2541 | * helper to solve this problem. It starts writeout of the truncated | |
2542 | * part in case it is in the committing transaction. | |
2543 | * | |
2544 | * Filesystem code must call this function when inode is journaled in | |
2545 | * ordered mode before truncation happens and after the inode has been | |
2546 | * placed on orphan list with the new inode size. The second condition | |
2547 | * avoids the race that someone writes new data and we start | |
2548 | * committing the transaction after this function has been called but | |
2549 | * before a transaction for truncate is started (and furthermore it | |
2550 | * allows us to optimize the case where the addition to orphan list | |
2551 | * happens in the same transaction as write --- we don't have to write | |
2552 | * any data in such case). | |
c851ed54 | 2553 | */ |
7f5aa215 JK |
2554 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2555 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2556 | loff_t new_size) |
2557 | { | |
7f5aa215 | 2558 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2559 | int ret = 0; |
2560 | ||
7f5aa215 JK |
2561 | /* This is a quick check to avoid locking if not necessary */ |
2562 | if (!jinode->i_transaction) | |
c851ed54 | 2563 | goto out; |
7f5aa215 JK |
2564 | /* Locks are here just to force reading of recent values, it is |
2565 | * enough that the transaction was not committing before we started | |
2566 | * a transaction adding the inode to orphan list */ | |
a931da6a | 2567 | read_lock(&journal->j_state_lock); |
c851ed54 | 2568 | commit_trans = journal->j_committing_transaction; |
a931da6a | 2569 | read_unlock(&journal->j_state_lock); |
7f5aa215 JK |
2570 | spin_lock(&journal->j_list_lock); |
2571 | inode_trans = jinode->i_transaction; | |
2572 | spin_unlock(&journal->j_list_lock); | |
2573 | if (inode_trans == commit_trans) { | |
2574 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2575 | new_size, LLONG_MAX); |
2576 | if (ret) | |
2577 | jbd2_journal_abort(journal, ret); | |
2578 | } | |
2579 | out: | |
2580 | return ret; | |
2581 | } |